I think that some posters here are losing sight of the fact that when you are measuring RPM there is something turning. Most likely that something is big, heavy and has lots of rotational momentum. Thus I say that if there is variation in the readings it is due to factors like quantization error (sync error in counting gate versus sensor time) or errors casued by poor sensor placement. If you make the corrections to fix these errors there is no reason that the averaging will work very well.

I also feel that the PLL idea (if you can design a good PLL circuit) should track well too. PLLs can easily track changes in speed of a rotating mechanical contrivance. I have had extensive expereince using PLLs in the past to track the data off a floppy disk drive where the data rate see was modulated by the speed variation of the disk motor.

Michael Karas