Hmmm, I think I found the source of my grief. It's not the anemometer, it's the fan.

From when I first turn the fan on:
After 1 min => 1375rpm
After 2 mins => 1385rpm
After 5 mins => 1400rpm
After 10 mins => 1405rpm
After 20 mins => 1405rpm
After 30 mins => 1405rpm

Thing is, every 30rpm equates to around an extra 1CFM through the radiator, and I started the fan up in the above test only after a 10 minute cool-down, so if it had started off really cool, the differences may be even larger. When I'm measuring stuff, a 50rpm difference which might have been occurring is a big deal.

It seems that in order to take accurate readings, I need to allow the fans to "sit" for ~10 minutes at any new setting before trying to take a reading.

The calibration stuff is all still okay though, 'cos I was holding a fixed RPM by ranging the voltages, but for the tests I was just trying to do, I was just setting a voltage and taking a measurement, and then set another voltage, and then I went back to the first voltage and noticed that the fan was now performing better. Now I understand why. The fan bearings must warm up to a stable operating temperature first.