I think it was Gordon McComb who responded in a recent thread something to the effect that "unfiltered PWM works better" which started/continued a heated discussion on the topic. I just found it:
In the thread "Control electronics power - need for isolation?" Gordon McComb wrote:
I wanted to make the following comment in that thread, but my memory of reading about the model railroad speed controller was too hazy. I think it's come back to me well enough now to post it:
ISTR a Railroad Modelling magazine with an ad or review for a power supply/speed controller which had as one of its features an unfiltered/pulsing DC output. The reason given was that at low voltage the engine would sit there because the power wasn't enough to get over the starting friction in the motor. The pulsing voltage gives a higher peak voltage and current into the motor, starting it turning at a lower average voltage than pure filtered DC voltage, and so the speed would be controllable down to a lower speed. For the ESC's, I presume the pulses of the lower-frequency PWM are long enough to get the motor turning at low duty cycle, whereas a higher frequency PWM at the same low duty cycle would appear to the motor as a low DC voltage, too low to overcome static friction. I don't know how PWM frequency would affect efficiency, but that's apparently not the only consideration. If it's important that your motors run and their speed respond somewhat linearly to PWM at a very small fraction of max speed, you apparently need to look into this. Or have a speed detect on the motor used in a closed-loop operation - but then if friction is too much, it may go into a jittery start-stop operation at low speed (the control loop overshoots to compensate), and lower frequency PWM control may solve the problem.