I've just received information from Dave Gravel at Newton forum, that the blocking on edges problem with rolling objects has been solved in version 2.02. Yeeehow!
on the other hand play with parameter of NewtonUpdate(), for example try NewtonUpdate(nworld, time_step / 64);
Then, to achieve the same velocity impression, I'd have to multiply it by 4 ( 64 / 16 = 4 ), Because inaccuracies in physics engines are higher if the velocities are higher, the effect then would be exactly the same