Hm that's weird. Then you probably have to use two models, one invisible model that does the collision detection (let's say with scale_z = 0.9) and one visible model hat is placed at the same position (with scale_z = 1) or a little below.
EDIT: Btw. I'm using an invisible player model for my game, too, that does not even rotate, but the visible, animated "clone" model does, and I'm not using the physics engine. It's a perfectly legit approach.
Last edited by Superku; 05/13/12 13:46.