Hmm, that makes sense, but if the only problem was acceleration, then it should put out the correct number of pulses, just very fast and it isn't. I put a nice counter on the Z axis step line and did a helical ramp and the count value was consistent but wrong. I would also think that in the case of an acceleration bug, it would "skip" steps only during the initial acceleration and I see consistent skips throughout the entire path, for example during thread milling.
During normal linear movements the counter indicates that mach is dead nuts accurate, during a helical ramp with all acceleration set to very safe and identical values, it is off by a thousandths or two, with acceleration of the ramp set to a different value, it has massive problems... surprisingly, if the arc center is (0,0) it behaves itself which is why I never caught the problem before starting a production run.