This sounds eerily familiar
If someone is working on this problem, they should also consider that this behavior is present on, or is triggered by the A axis.
Under certain conditions, if the A axis and the X axis (for example) start moving together, MACH ignores acceleration and just slams the X axis from zero to feedrate. I don't recall off hand the circumstances, but I posted some details on this quite some time ago.
Sounds like this may be a symptom of the same bug, so I thought it worth mentioning.
I've had issues with the A stalling in a similar fashion, rates were WELL below anything that would cause the axis to stall. Interestingly enough I found that reducing the rates themselves in the program did nothing to stop the stalling behavior, but when I backed down the FRO even just a tiny bit, it would no longer stall.
The problem, as I understand it, affects "secondary" axes, and helical moves. If you have one or more axes in motion at a reasonable speed, then do a helical move that brings in a "secondary" axis, Z in my case, the CV code does not correctly account for the acceleration capability of the secondary axis, resulting in the primary axes entering the helical move too fast, thus requiring the secondary axis to accelerate essentially instantaneously. Apparently, it's not that the correction is not being done correctly, but rather that it's not being done at all! The code simply wasn't there. The same problem apparently existed until fairly recently on linear moves as well. Brian fixed that a while ago, but never got around to making the changes to the helical move code. He know exactly what to do, and how, so he should have a fix very soon. I'm actually surprised this hasn't been a much bigger problem for more people. I suspect once the new code is out, a number of problems will vanish, and many of us may be able to kick up our velocity and acceleration settings closer to the limits.
Regards,
Ray L.