There is a lot of theoretical discussions on that issue here:
http://www.physicsengine.com/forum/viewtopic.php?t=748
http://www.newtondynamics.com/forum/viewtopic.php?t=1051&highlight=timestep
Tomek explains the issue like this in the Newton Forum:
--------------------------------
If you pass more than 1/60sec, then Newton will clamp it to 1/60, so it's first need for your own time-slicer.
Second one is, that normally you want not only to step physics, but also do your own 'thinking'(for example AI), and you either run in at physics rates, or at much lower rates, depending on game genre.
Therefore, you need some kind of 'time-slicer' for the rest of subsystems too.
Besides, if Newton internally slices time period passed, you cannot do anything between iterations...
But of course if you don't mind these constraints you can rely on Newton slicer
--------------------------------
BTW, I have updated the CarLab to version 1.0c with a lot of new functions and also now using the 1.32 version of the Newton system. Should work more stable on Celeron processors due to different compiler optimization.