I agree, I think the major difficulty is modeling the aero. Aero is very non linear, so would require quite a big lookup table containing various speeds, rakes, rideheights, angles of attack, etc. This can not be implemented in the current physics sheet of rf2, if I am correct.
I do think that the mechanical part (suspension kinematics, dampers, drivetrain) can by modeled and simulated quite acurately. Traditional commercial vehicle dynamics software focusses on this as well. This might also be the reason that simple non-aero cars get a lot of praise for their realism.
The tyre model in rf2 might be on par or even better than some proffesional vehicle dynamics software. rfPro probably the benchmark here.
Also I dont thinks s397 get all the info from the manufacturer to properly replicate a car. I wonder how realistic it would be if they could accurately get all the paramaters rf2 currently uses.
I'm very interested to know if there are smaller race teams or even OEMs that modelled their car and use rf2 for simulations, and what their conclusions are. It seems a powerfull piece of software. And very entertainable as well.