EMR doesn't really care about the fundamental frequency (within normal, sensible and practicable pwm frequency limits for such things, ie 100 to 100kHz) it cares about the RISE and FALL time of the signals edges. Hard switching your fets WILL produce a "squarer" output wave form and hence the harmonics of those edges will produce more EMR. IN all cases, as usual, it's a trade off between EMR and efficiency. The harder your switch, the less switching losses you get, but the more EMR you get.
For something like an ultrahigh efficiency DC:DC or similar, where you want to be able to drive the smallest inductor possible (for cost and space reasons) then you would choose a fundamental pwm frequency as fast as possible, (these days, up above 200KHz probably), and they would need to have hard switching, with times to fully cross the miller plateau below 1us to prevent the FETs going over temp.
But in your case your load is fixed already and it IS a huge inductor (probably, most high current DC brushed motors are massively inductive due to design constraints) so you know you can switch very slowly, so slowly as to almost make the switch losses irrelevant, because they happen so infrequently. Also with just 12v, your dV/dT isn't going to be terrible either, again, helping matters. I'd pick a decent FET driver (like the LT1158) and carry out a gate drive impedance exercise with the real fan motor you are driving to work out the lowest fundamental PWM you can use (before current ripple becomes excessive (need to do that at 50% duty - max ripple point) perhaps aim for <10% ripple current) and how slowly you can then drive the gates. I'm going to guess that probably even a very soft drive of say 5us to switch will probably be ok, because it's only happening once every 10ms!