After a fun debate over on the G+ 3D Printing group, I have been proven wrong (which I do not mind at all – being correct is far more important than being right, if you know what I mean). The discussion was regarding using SAE rods to drive the Z axis but choosing layer height based on nice round numbers of millimeters, like 0.25mm or something.
Turns out, that results in rounding errors, because for instance with a standard 18 TPI rod and a 3200 step revolution, 0.25mm (as an example) means 1128.888888888888 steps per layer. This is going to result in rounding errors, so each layer will not necessarily be exactly the same height. It sounds miniscule but when I tried it last night, it really does seem to make a small difference. The difference is magnified the smaller the layer height.
Moving to 0.254mm layer height seems to smooth it out quite a lot.
In reality it may be that moving to metric threaded rod is the best solution, then you have nice round numbers in the firmware for steps per millimeter.