Ah!
I think this part of the doc is relevant:
This means morphing from a microstepping to full-stepping when the motor speed is above where microstepping is of any benefit (above 4 revs / sec.)
Because it’s changing the motor step interval on the fly as the speed changes, then their definition of “accuracy” probably applies to the endpoint of the motion, not to the trajectory in between, which would explain why it produces odd results when the motion should be running at a constant speed.
We’ve seen a number of cases where “smart” / servo / closed-loop motor drives produce poor results, because they’re optimized for something other than laser control that requires precise speed control.
Here’s one horror story:
This seems like a similar situation.