I did some fairly serious experimentation with modified servo motors a few years ago to determine how accurately they could be controlled. After running a lot of automated tests and compiling the results I came to the conclusion that the control circuitry and algorithms employed were not compatible with accurate motor speed control, since they were intended primarily for stable position control, with accuracy of any kind being less important. The most significant point is that the controller implements "dead zones" in order to prevent jitter.
Here's a graph that I plotted of the results. The horizontal axis is the pulse duration in microseconds and the vertical axis is rotations per minute. I used three different servos for the tests, represented by different colours, all Hitech HS-322 if I recall correctly. Dead zones cause the saw-tooth appearance of the control versus speed graph. Also, and not surprisingly, the results vary according to the voltage supplied. As the batteries run down, the motors slow down.