In telephony the traditional human hearing response band is 300Hz to 3KHz. That is what telephone systems have long been designed to carry. My old level measuring set used 800 and 1600Hz as its test values, those defining the band that humans generally hear best. HiFi systems spread considerably each end, as do peoples abilities to hear higher frequencies. Lower frequencies rarely upset anybody, 50Hz corresponded nicely with the normal frame rate of RC systems and didn't need too much in the way of circuitry to work. Higher frequencies in the audio range will be more noticeable if the motor is susceptible to respond to them. Using even higher frequencies will depend on the ability of the hardware in the control chip to provide a useful output at that rate. Probably waiting for the day when PIC chips operate in the GHz range.
The basic problem is that any motor has the same components as a loudspeaker, and given a suitable signal, will respond like one. The noise either comes from the armature responding to the PWM signal by modulating its rotation with a vibration, or the windings moving on the armature. Full size motors, at the end of the winding process, get their armatures dunked in a vat of varnish (but keeping it off the commutator). When dried and set, they are relatively quiet. While this can be done on a model size motor, a problem arises in that a smaller mass needs less force to make it move, so model size motors are inherently more liable to respond to the control signal.