I think that in rF2 the self aligning torque too easily reverses direction with low caster values. There are many cars in rF2 where the self aligning force reverses direction... Yet, IRL even with pretty low caster angle rarely do you ever get a reverse in direct of the forces at the steering wheel. No doubt that the reverse in direct at a certain slip angle is technically correct in rF2 at the tyre, but there must be other forces IRL that keeps the feedback at the steering column positive.
Especially with power steering, why would a modern racing car ever have geometry that translates to a reverse in the direction of force feedback? Maybe there's a reason IDK.
Maybe there are a lot of cars that do reverse direction IRL, and I just haven't driven said cars, let alone pushed them at high front slip angles.
It might be desirable for the caster self-centering element of the FFB to be reduced near the optimum slip angle so that the steering forces sum to a force level for the driver.
I disagree that the Steering force of the Merc GT3 becomes negative at a certain point - instead, it just reduces near the optimum slip angle. If it was negative, and you let go of the wheel, the car would go full lock and spin instantly, which it doesn't.
If you take any 90's F1 car for example, and if you drive it with a Thrustmaster T300, you can easily man-handle the steering wheel under any condition. Whereas, the same car, but with a 1:1 nm force output on something like a simucube, the car is now man-handling you!
However if you reduce the mechanical trail by reducing the caster or moving the linkages rearward, the contact patch warps, and the center of forces of the tyre sometimes moves slightly in front of the lead point. The more the tyre warps with steering lock and load, the further forward the forces move - this reduces the centering force. Good geometry will not linearly increase in force as the steering lock increases for a few reasons, but from the drivers perspective, you'd never feel understeer, and the steering would be heavier than necessary.
When you have a car which will quite happily numb your limbs after 20 minutes, calming the centering force under high G cornering is really good idea. With a 1:1 setup, these 'low caster' cars don't feel weak at all, they just have a zone near the optimum slip angle where the driver can be more sensitive to the steering wheel, instead of trying to fight it.
The best example I know of is the Formula Renault 3.5. On a Thrustmaster, it feels very lack-luster and doesn't really make any sense. But with a big DD wheel with 1:1 force, it actually feels like very purposeful and communicative suspension geometry, designed to allow the driver to get the absolute maximum out of the car under high G cornering, and to not be fighting it.
Same with the Merc - it just feels like it's understeering on a Thrustmaster, but on a big wheel it is actually really nice.
Here it is: 
