Notes on Hirose's KR-II
I think the limit is when the wheel on a given segment goes over a bump, where the worst case would be a vertical step.
A bit of calculation shows that for a step of height 0.5 r, the velocity is about 1.68 times the forward velocity.
h=1-sqrt(1-x^2)
v = -x/sqrt(1-x^2)
If the step is higher than the diameter of the wheel (or, in fact greater than half the wheel diameter), forward motion would have to stop while the wheel "rolls" up the face of the step. Given that obstacles aren't always nice steps perpendicular to the direction of travel, this effectively limits the obstacle size to the wheel radius. Some form of predictive motion might help, but that would require a fast sensor (however, there are optical sensors that might work?).
One could also suspend the wheel from a spring loaded trailing arm type mechanism, so that it can swing back and up. This would also give a feedback for the down force.