Julio, your idea seemed to be what I was looking for in the first place, but even changing it to that way won't make it work much better, and I reall< don't know why.
I now don't use the convex hull anymore but instead create some chamfers to represent the track and attach them by using a limited slider joint to create a suspension, cause otherwise the tank hops around too much. See this shot how I do it :
(Ignore the convex hull for the thread. It's still there but I have disabled collision between it and all other objects, I only use it to get the position for rendering the tracks at the right place)
So localizing the contacts with the chamfers *seems* to work, but if you take a look at the upper left corner you'll see the different number of contacts for each side of the track (the numbers are so high cause I'm using accumulative time-slicing, and I count the contacts generated by all NewtonUpdate calls, so don't worry about that). So it's almost the same as with the old version that used a convex hull for the track.
So do you know what's causing the different number of contacts, even with this configuration that basicly shouldn't create a different number of contats? Due to the suspension I'm using for the chamfers they all touched the floor in that screenshot, but still different contact count for the two tracks.
If you need a demo for testing/debugging, drop me a line and I'll try to get it up tomorrow. In the meantime I'll try to tweak around a bit more.