your problem is that the step rate specified in the Mach 3 XML is related to the machine clock speed - not the value you entered on screen, so it is likely MACH3 is not sending wide enough pulses and so your driver is missing some step pulses under 3-axis load. Go back into the MACH3 setup, and change the pulse width setting to 5 or 6 and resave the XML... it might be worth checking the MACH3 kernel speed settings...