I've been folowing this thread off and on and I just want to throw an idea into the pot.
It seems to me that there may be a conflict between reading the RPM only once per rev and the built in power compensation in some variable speed controllers. I have an ammeter on the output of my Minarik brand drive and I can observe the power changing yet the RPM is (relatively) consistent.
For example, when the tool first starts cutting, the spindle will slow down. Then the variable speed conrtoller will compensate and throw more power at it.
If this all happens within one rotation, MACH has no ability to measure the actual speed that the spindle is turning at the END of the rotation since it only has available the time taken to complete one rotation. i.e. Mach would measure a lower speed since the initial slow down would be included in the total revolution period.
What I'm thinking is that Mach would adjust for the speed it thinks the spindle is turning based on the avilable info, namely the time interval of one revolution, when the spindle is actually turing at a higher speed becuase the controller has compensated and increased the power.
This scenario would self correct over a few revolutions as the rpm stabilized (assuming a relatively consistent stock density) during the cut. This seems to be consistent with the behavior people are experiencing.
If this is the case, more readings per rev would be a potential solution. In this way, Mach would become aware of deviation within one revolution which have been corrected within that revolution.