Something tells me that the simulation frequency vs. display frequency ratio is an important aspect, as in, sampling theorem, aliasing artifacts and such, which may surely be a 'nice' contributor to perceived jerkiness of movement.

After all, changes in position over time, generated by the physics engine, is a signal, which is output by the physics engine, and sampled again by the graphics engine, at a possibly different frequency. Now if the resampling method used is not good, the end result, i.e. graphics output, will have artifacts.
From my experience I'd say, as others have noted, odd ratios like 67 Hz : 60 Hz, make simple resamplers produce especially bad results.

Upping the simulation rate, i.e. samplerate of the "source" (physics engine), certainly is one approach that works, but it's a low-tech one, imposing high computational cost - such goes the common wisdom in DSP land I think. (I'm no expert, but dabbling in that sort of stuff recently).
There would have to be some sort of filtering / interpolation going on as an alternative.