The most important determinant of how an AWD car handles are the differentials. Every other aspect of the setup is secondary.
Why this is the case - and how to change it - is one of the most difficult problems in vehicle dynamics. Because of this, and because there is so little literature on the subject of AWD racecars, the effects of the differentials on handling are often poorly understood or not given enough weight when trying to set up an AWD car.
As is often the case, it's just not this simple. We're going to have to make a lot of simplifications and assumptions, and as always, real-world testing trumps anything you read on the Internet.
What Makes AWD Diiferent from FWD or RWD in Handling Terms
Unlike a FWD or a RWD, each end of an AWD car is not independant. When attempting to set up a RWD or FWD for optimum handling, it is convienent to assume that changes made to one end of the car do not affect the other end. This isn't exactly true - lateral weight transfer, for example, affects the opposite side corner as well - but as a simplification, it works well, as the crosstalk between the two ends of the car are usually complementary.
For example, to reduce understeer on a FWD car on corner exit, one approach is to increase the amount of rear weight transfer distribution by softening the front bar, stiffening the rear bar, or both. One would expect this to increase the amount of load on the outside rear tire (at the expense of the load on the inside rear tire) and thus reducing the amount of grip on the rear of the car, promoting oversteer. Often overlooked is that the increased outside rear load will increase the load on the opposite diagonal (on the inside front) which, all else being equal, should increase the amount of front grip, which in turn promotes oversteer.
In effect, "unsticking the rear" actually does two things - unsticks the rear and sticks the front - but both changes work together to move the handling balance of the car in the same direction, so we can pretend that the ends of the car are independant. So long as the only crosstalk mechanism is weight transfer through the sprung mass, this assumption holds.
From this, we get the usual methods of changing handling balance - less weight transfer distribution on one end of the car causes it to stick more; more weight transfer distribution on one end of the car causes it to stick less.
But this assumption only holds if there is no other crosstalk between the ends of the car other than weight transfer. On an AWD, this assumption is NOT true. Each end of the car is connected to the other by the driveline, and the nature of that crosstalk is determined by the differentials. This is further complicated by the fact that each side (on both ends!) of the car is also connected to each other by the driveline and a differential, and these differentials are connected to the centre differential in series. Not only do the differentials affect the handling of the car, they also affect each other - which, in turn, alters the handling of the car.
The ultimate handling balance of an AWD car is thus a complex interaction between the differentials (first amongst themselves, then modified by weight transfer) weight transfer distribution, static weight transfer, and ultimately the tires.
A Bicycle Model of an AWD
To get our feet wet, let's first pretend that the car is a bicycle, with no front or rear diffs. This lets us concentrate on the centre diff for the time being.
Imagine the car in a steady-state, maximum-G turn with a neutral attitude (ie, both front and rear tires are cornering at the limit, with no reserve capacity, meaning in turn that there is no latent understeer/oversteer in steady state cornering)
In order to exit the turn, the driver must unwind lock from the steering wheel. This reduces the amount of cornering force delivered by the tires, and creates reserve capacity in the tires which can be used to accelerate the car.
In a RWD, if the driver applies throttle, the reserve capacity in the rear tires is now taken up by the acceleration force. Simultaniously, weight is transferred rearward, increasing the total grip capacity of the rears while reducing the grip capacity of the fronts - which is OK, because unwinding the steering wheel has created reserve grip capacity in the fronts, so we can afford to lose a little because the fronts are not being fully utilized anyway. This increase in rear grip due to rearward weight transfer allowss more throttle to be applied, which in turn creates more rearward weight transfer etc. Assuming we have an excess of power, we either eventually reach the ultimate limit of the rears (in which case we have power-on oversteer) or rearward weight transfer reduces the grip of the fronts beyond the reserve grip capacity created by the reduced steer angle, and we have power-on understeer.
In a FWD, applying throttle uses up the reserve capacity in the fronts and at the same time, rearward weight transfer steals grip from the fronts. Thus, the harder we accelerate, the more grip we lose from the fronts. This means FWDs will always understeer under power.
In an AWD, applying throttle uses up reserve capacity in both the fronts and the rears. If we assume a centre differential with a 50/50 torque split, both front and rear are using up reserve capacity at the same rate as throttle is being applied. But because rearward weight transfer is stealing grip from the fronts and giving it to the rears, the fronts will run out of grip first and the car will understeer. Assuming a sufficiently sophisticated centre diff, the diff will "notice" that the fronts have run out of reserve capacity and route power rearward - further increasing acceleration, which in turn increases rearward weight transfer and with it, understeer.
It is the essential nature of AWD cars with 50/50 centre diffs to understeer under power
Now, let's examine what happens if a setup change is made to try and reduce rear grip and promote oversteer:
Assume that some setup change has been made that reduces the amount of rear grip availible. Commonly, this is attempted by increasing rear weight transfer distrubution, but we're discussing a bicycle model at the moment, so we can do it via tire pressure of reducing the width of the rear tire or with a harder compound or whatever.
If the reduction of rear grip is slight, the fronts run out of grip first, the car understeers, the diff sends power rearward, and the rear runs out of grip sooner than it would have. The balance remains understeer (although it may be oversteer on corner entry and midphase) but now the car cannot accelerate as hard as it could previously. The on-power balance does not change, but the car is slower.
If the reduction of rear grip is drastic, the rear may run out of grip before the front. The diff, however, will "notice" the loss of rear grip and send power forward to the fronts, using up the reserve capacity in the fronts, and subsequently increasing rearward weight transfer and increasing (marginally) rear grip. Balance in this case is hard to typify - the rear will be sliding around a lot, but neither will the car tolerate any extra steering lock. Depending on the instantanious situation, the car may be said to oversteer or understeer. It may actually oversteer somewhat, but the diff sending power forward will tend to reduce the oversteer.
Put another way, this setup on a RWD would have diabolical oversteer, but on an AWD, the tendancy of the diff to increase understeer will tame it down a lot - under power. That latent oversteer is still there however, and depending on how extreme it is (and it takes extreme latent oversteer to create ovesteerishness under power) that oversteer can manifest itself on corner entry/midphase.
The net result of trying to create power-on oversteer on an AWD with a 50/50 diff is a car that likes to occasionally snap-spin on corner entry or midphase. It makes for a very difficult and unforgiving car, especially in transitions or the rain. The car will also be slower than a car set up to maximize grip. And you still probably won't get the car to exhibit power-on oversteer anyway
The only way around this is to use a diff that has a rearward-biased torque split. With such a diff, the power sent forward will stop when it reaches the forward bias percentage. Assuming that this bias percentage is reached before the fronts run out of reserve capacity (and assuming that rearward weight transfer never robs this reserve capacity below the point of the bias percentage) the car will now behave as a RWD, which can mean power-on oversteer. There is no other way to get power-on oversteer without compromising the performance of the car.
Adding the Front and Rear Diffs to the Equation
The bicycle model makes the assumption that each pair of wheels on each end of the car reaches their limits simultaniously and that the limits do not change (except for longnitudnal weight transfer). Unfortunately, this is rarely the case.
If something happens (on either end of the car) to reduce the amount of grip at that end, the centre diff will "notice" and route power towards the other end of the car. This tends to mitigate any tuning aimed at "unsticking" one end of the car (which, in turn, can make it seem that the car is insensitive to balance changes - it is)
The exception is if something interferes with the actual power transfer mechanism itself - typically, the interaction of the centre diff with either the front or rear diffs.
Imagine the case of a car with a limited-slip rear diff, and an open centre and front diff. The nature of an open diff is to route power to the wheel with the least amount of grip. On corner exit, the inside front is very lightly loaded and has very much less reserve capacity than the outside front. Exceeding that reserve capacity with power application spins up the inside front, and 100% of the torque sent forward is now used in spinning the gripless wheel. But the centre diff "sees" the gripless wheel as well, and it sends 100% of its power to the front - and the car goes nowhere.
If all three diffs are open, then a loss of grip on any one wheel renders the car immobile. This is why all DSMs come with a limited-slip centre diff.
If the centre slot is limited-slip, the centre diff will "notice" the loss of grip from power on that end of the car going out the spinning wheel, and will send power to the opposite end of the car. This never quite reaches 100%, both because of the nature of the locking mechanism, and because if 100% power goes to the other end of the car, the spinning wheel must stop spinning (it no longer sees any power) and when it restablishes itself, the centre diff will "notice" and send power forward again.
Because of the inherent understeer in an AWD, it is almost guarenteed that the inside front tire will reach its limit first. The consequences of this on balance are indeterminant. On the one hand, the inside front is no longer contributing any grip (either laterally OR longnitundally) which should tend to promote understeer. On the other hand, the more heavily-laden outside front is not being sent any power, so it should have reserve capacity left for cornering, which whould tend to reduce understeer. Which characteristic will dominate the balance of the car is tough to determine.
Two things can be determined though:
Firstly, the car will not accelerate as hard as it could were power not being sent up in smoke through the inside front.
Secondly, when some mechanism stops power from being wasted and grip is re-established to the inside front (through the centre diff sending enough power rearward, or the more prosaic method of he driver backing off the throttle) the sudden restablishment of the contribution of the inside front to cornering force tends to suddenly jerk the nose of the car towards the inside of the turn - unsettling for both car and driver.
From this, it should be clear that we don't want the inside front acting as a "power fuze" - we need to make sure we limit slip on the inside front and make sure (as best we can) that both front wheels reach their limit simultaniously. That means we need a limited slip differential in the front slot as well as the centre slot.
The same holds for the rear as well, although the rear winds up living a much easier and happier life than the front. Because the rear is always the beneficiary of weight transfer under acceleration, it is always going to have more grip than the front - that's the nature of inherent understeer - and so the inside rear is less likely to break loose than the inside front. As power levels rise, the rear diff starts to become increasingly important, and if a rear-biased centre diff is fitted it gets even more so, but in the majority of cases with tires of any decent grip level, the rear diff plays the smallest part in determining handling.
As always though, There's No Such Thing As A Free Lunch. While installing limited-slip diffs in all three slots both reduces the amount of power-on understeer and increases the amount of acceleration out of the turn, there is a tradeoff. Unless the diffs are purposely set up to only function in the applied-power mode (ie, not under braking) they will tend to increase corner-entry and mid-phase understeer. This is the nature of the limited-slip diff.
In the larger scheme of things though, the trading of corner-entry understeer for reduced corner-exit understeer normally turns out to be a good one, as without the limited-slip diffs, corner exit performance is so badly hampered (especially as power levels increase) that the better turn-in from not having the limited-slip diffs doesn't outweigh having better exit performance. Given that the real advantage of AWD over other driveline layouts is the ability to put more power down (given the same size contact patches) than either FWD or RWD, compromising power-on handling just doesn't make any sense - and there are ways to reduce the corner-entry understeer to manageble levels.
A good compromise setup is to not fit an agressively locking diff in the rear slot. As the rear end is the least likely end to break loose under power, it is the least likely to need help from a limited slip diff, and so if a looser diff is fitted in this slot, corner-entry understeer can be reduced.
A Discussion of Locking Mechanisms
So then, it can be seen that what we need is a diff that can send exactly as much power as that contact patch can stand - and no more - to the appropriate corner of the car. The centre slot gets to route power front/rear, and each end slot handles side to side.
The best way to do this is with some sort of computer-controlled hydraulic locking mechanism, loosely related to a modern automatic transmission controller, but with the ability to do degrees of lockup (rather than just on/off as with the band clutch actuators in an automatic) Easy to say, tough to do. This winds up being rocket science. Both Formula 1 cars and World Rally Cars use this type of technology. It is effective and completely worth the effort. It is also horrendously expensive and difficult to tune. A typical WRC transmission, including control systems, is a milion-dollar part.
There's a reason why those WRC cars are able to handle such extreme yaw angles and slip angles, and yet never seem to slow down. That's not driver (although the average WRC driver is certainly no scrub) it's technology.
A glimpse into the kind of complexity needed to do computer-controlled AWD power routing correctly can be seen by the number of inputs fed into the controller. There will be 4 wheel speeds (as one would expect) but the controller will also get information from brake pressure sensors, accelerometers, yaw angle and rate sensors, steering angle sensors, and even a sensor on tha park brake position.
There's just no way a mechanical system can duplicate this kind of performance, and no way that mere mortals can get their hands on this kind of technology (or the expertise needed to calibrate it). What is needed then is a mechanical system that best approximates what a full-bore computer system would do in similar circumstances.
The locking mechanism that comes closest will put down the maximum amount of power with the minimum amount of understeer - remember that it's always going to understeer if the torque bias is split 50/50
Spool The spool is the simplest locking mechanism possible, being a solid shaft with no differential mechanism at all. Because it is locked solid 100% of the time, it forces whatever slot it is installed in to rotate at the same speeds on both sides. It is simple, lightweight, and very strong. It also causes horrendous understeer and (when used in the centre slot) can lead to a kind of "hop" as all four wheels fight each other. Typically only suited for cars that never turn (drag cars) or perhaps for use on loose surfaces.
Viscous The viscous coupling consists of a pair of shear plates running in close proximity inside a can filled with a silicone gel of very high viscosity. The gel is resistant to shear, and tries to oppose any difference in speed between the two plates (one of which is connected to each side of the differential) If enough shear action takes place, the fluid heats up, expands, and locks the plates together nearly 100%
This two-stage action makes for a differential that is very progressive, requires almost no maintainence, and can provide excellent performance - I have once (by accident) run a viscous centre diff on a chassis dyno in FWD mode, and it actually locked solid. It is also the diff type fitted as OEM in the centre slot, and optionally in the rear slot.
Its major drawback is that it takes a finite and noticable amount of time before it starts to really lock up - on the order of 1-2 seconds. The givaway characteristic is what sounds like clutch slip under power - step on the gas, the engine spins up, then slowly starts "grabbing" and the car starts accelerating faster (if you've never experienced this in a DSM with the OEM centre diff, you're not getting on the power early enough or hard enough)
Furthermore, particularly on a car with a lot of power, the transition from "viscous" to "locked" can be pretty violent - especially if steering lock has been added in to combat the understeer that happens while the diff is spinning up, which can jerk the nose into the turn pretty harshly. It's a wonderful part for a street car, but doesn't work very well in a racing environment.
This may well be the kiss of death for the Cusco "tarmac spec" centre differential, which has a rear-biased torque split - currently (Nov 2004) the only diff availible for the DSM that is not a 50/50. Amazingly enough, this diff does not come with any sort of auxillary locking mechanism; it is an "open" diff. That means it falls back on the OEM viscous coupling (which is not co-located with the actual differential) as its locker, which means that it is prey to all the drawbacks associated with the viscous.
The only National-level competitor using one is doing so in an ESP car, which means he is (comparitively) down on power, as the ESP rules limit him to the factory T25 and forbid any sort of boost controller. As such, he is at much lower risk of spinning up the inside front (with or without a front LSD) and running into the time delay penalties associated with getting the viscous to lock, or with the abrupt handling changes associated with the viscous locking after a period of slip in the diff. I would expect these problems to become more prominent with increasing power levels.
I haven't personally driven this car (or one similarily equipped) but from the exterior, it is neither dramatically faster nor dramatically slower than other DSMs, nor is the cornering attitude visually different. How much of this is due to the diff, other aspects of the setup, or to driver technique is indeterminant. Call it an open question for the time being.
The only marked tendancy this car has displayed is one towards breaking rear axles, instead of front axles as is the normal DSM trait. Obviously the diff is doing something, but the jury is still out on if it is a net positive or not.
The exception may be in the rear slot. The viscous doesn't contribute much to corner-entry understeer, and the time delay between partial-lock and full-lock doesn't appear to affect the rear very much, if at all. An OEM viscous in the rear seems like a good choice.
Clutchpack The clutchpack diff uses a series of friction plates and springs to provide a locking mechanism. By varying the spring force and/or the friction characteristics of the friction plates, one can change the amount of lockup provided at various differential axle speeds.
This is a tried and true solution with decades of history behind it. It works. There are, however, some characteristics of this design that limit how effective it can be at approximating the ideal lockup forces we'd like it to provide. Firstly, it tends to be very linear, in that the amount of lockup provided is a linear function of the friction/spring preload characteristics of the clutchs and springs used. As the "ideal" lockup curve is not necessarily linear, that means this diff design will, at some parts of its operating range, induce more understeer than is ideal. Secondly, the clutches and friction surfaces wear out with time, changing the slope of the power transfer curve. In order to get best performance out of the diff over its rebuild shift, it may have to be installed "tighter" than otherwise optimal with an eye towards it "loosening up" as it wears. Thirdly, one can expect a good deal of "trial and error" tuning, it which the diff is reconfigured and reinstalled with different values of locking agressiveness in the search for the best compromise setting. This is likely to be labour-intensive.
"Phantom Grip" The "Phantom Grip" differential is the poor cousin to the clutchpack. It works in exactly the same manner. The difference is tha instead of installing a dedicated set of clutches and springs, the Phantom Grip installs a spring-loaded metal block that rides along the nose of the spider gears in the differential, using the metal-metal contact between the block and the gear as its friction surface.
Aside from the dubious wisdom of placing a signifigant sideload on a gear not designed to take it, the Phantom Grip's main problem is the relatively small friction surface (which limits the amount of locking possible) but also one of wear - as the block and the gear lap into each other the amount of friction between them will drop, along with locking potential. This is a band-aid solution at best.
Torsen (Quaife) The Torson differential works via a special arrangement of gears inside the housing, which in a nutshell exploits the fact that worm gears like to be driven in one direction and not another - don't ask to explain in any more detail than that, as I don't rightly know myself.
Because this is a 100% mechanical solution with no wear surfaces (aside from normal gear->gear wear) this is a very progressive and consistant solution that provides easily predicable handling response. Furthermore, the Torsen is unique in that it reacts to torque differential, not speed differential as do other locking mechanisms. It is VERY good at sending only the amount of torque that a contact patch can stand to that corner, and it reacts very quickly and smoothly. It is probably as close to the "ideal" torque transfer solution as can be had out of a mechanical system.
It does, however, have a couple of quirks. The first is that it is only availible in a 50/50 torque split (and that may or may not be a consequence of the design, I don't know). The second is that it requires a little bit of resistance from the inside wheel to "push against" before it can start locking. The net result is that unless there is a secondary locking mechanism installed in parallel (like a clutchpack or viscous) if a wheel comes off the ground the Quaife in that slot becomes an open diff.
In the centre slot, this is no big deal, as it is unlikely that a DSM will ever lift BOTH rear or front wheels clear of the ground, and even if it does, the OEM viscous is retained as part of the assembly packaging. In the front slot, this is a non-issue on a 2G DSM as there is no reason to be lifting the inside front wheel off the ground on corner exit. With a 1G (or any car with a MacPheason strut front suspenion) it would be nice to b able to accomodate the limited camber gain in roll from the strut by using 100% weight transfer on the front and setting the static camber such that the fully loaded outside front was in its camber "happy place" at full roll - but to do that means lifting the inside front clear of the ground, and a Quaife in the front slot would "go open" and very much upset the car. No reaserch has been done to determine if a 1G (or EVO, for that matter) is better off compromising ultimate front grip for better torque transfer characteristics (by using Quaife front diff and less forward roll resistance distribution) or by sacrificing torque transfer but picking up ultimate grip (by using a clutchpack front diff and more forward roll resistance distribution) - and the answer may very well be tire-dependant.