[ODE] Another strange wheel rotation bug

[Remko Jacobs (act-3d.com)]

I fixed the 'bending wheels' problem using the following methods. The   
problem still exists but at least you can reduce it to acceptable levels:

 - Decrease the step size (update of 200-300Hz)
 - Different CFM and ERP values
 - Different masses
 - Bigger wheels (The bigger the wheels, the slower they have to turn for   
the same speed)

I'm experimenting with a driving simulation and the results are stunning!   
However, after some experimentation, the car started to bounce and the   
handling was completely gone.

It turned out that the wheel is not rotating around it's axis but around   
a point slightly above that. Gravity was causing this offset. Gravity   
pulls the wheel down so the rotation axis of the hinge is too high. What   
really surprised me, is the solution I found:

When I set the gravity to zero the problem does not occur (as expected).   
When I crank up the gravity while the simulation is running, the bug does   
not come back but the wheel rotates around the correct point. Does anyone   
has an explanation for this? Why the rotation screws up when I initialize   
with gravity?

 -Remko