It all boils down to the waveform you want. You are going to have to look at the trapezoidal pulse you want in your square wave and decide what you want it to look like. How much of the rise and fall slope do you want in each pulse relative to the steady state "full-on" part of the pulse? Balance this with your desired frequency to get acceptable power dissipation. THe worst case scenario is the square wave at the smallest non-zero duty cycle that you want, at the highest frequency that you want. Remember- this is the WORST case scenario. Whatever numbers you specify here, you will get a more ideal square wave at any lower frequency or higher duty cycle.
THis means that your 10% transition time criteria is very strict if that is what you want at a worst case scenario. When I was doing calculations, I too used 10% as my desired transition time per pulse in the worst case scenario (10% the on-time of the shortest pulse). But after doing the heat dissipation calculations I realized it was not going to be possible without very large fast transistors with very powerful gate drives. It became a lot more reasonable when I used something like 66% transition time for the worst case scenario (2/3ths of the shortest pulse is transition and only 1/3 of it is steady state-full on.). This meant it would not be a very nice square wave at low duty cycles and it would be lossy...but guess what? At low duty cycles it also dissipates the least power so it wasn't a problem. At higher duty cycles, the on-time got longer but the transition times stayed the same, so the square wave got much nicer as the duty cycle (and power dissipation) increased.
It has very little to do with how fast the MOSFET can switch. You can always make the MOSFET switch faster by using a higher current gate driver and it can almost always switch your MOSFET faster than you actually need it to (making problems like RF, EMI, and ringing).