They may or may not have sufficient filtering in there for whatever PWM signal timebase you can generate.
As a guess, they probably would put some filtering that would handle mains ripple on the input line so if you had a PWM timebase in the kHz it might have low enough ripple to not cause issues.
Worst case will be 50% duty cycle and I would expect the effect of too much ripple would be a beating of frequencies (sampling vs. PWM) resulting in a modulation of RPM with time.
Since you have to generate the PWM anyway and likely have to boost the voltage from logic level to 10V, you may as well filter it and buffer it properly, in my opinion. The few parts (say an LM324 and a few resistors and capacitors) cost virtually nothing compared to playing around with this kind of thing.