Kai's method is clever, but engineering on the principle of "good enough" if you have a very long period and time 100 pulses in - that takes you 100 Seconds, with a 1 pulse/rev vane, but reduces your uncertainty to 1 in a 100. Add a 10 pulse /rev vane and that would be a 10 seconds response for a step change in speed of 1 RPM, which may be acceptable.

At the top end, you have 1 pulse in 1.5 msecs, 10 in 15mSec, which you can time to 1usec resolution, and you have 15,000usec of resolution from a timer, so there is easily enough resoltion left at the top end too.

Steve