From Wikipedia:
"Although the long term effects due to ultrasound exposure at diagnostic intensity are still unknown,[24] currently most doctors feel that the benefits to patients outweigh the risks.[25] The ALARA (As Low As Reasonably Achievable) principle has been advocated for an ultrasound examination — that is, keeping the scanning time and power settings as low as possible but consistent with diagnostic imaging — and that by that principle non-medical uses, which by definition are not necessary, are actively discouraged."
Another interesting one is phase noise and distortion and crossover distortion (if you use a simple class-B amp) means you can send a sine wave into a high power amp, but the amplified noise level at audible range might be icky. As a personal anecdote I've noticed this with commercial ultrasonic cleaners. Sure... noise level might be 30 dB down at audible range, but pump a 100 watts at 40 KHz into it and I'm listening to 0.1 watts of noise at audible freqs which is quite loud indeed. Theoretically ultrasonic cleaners should be inaudible, but high powered ones are not. Much as intermod distortion is hilarious with high power RF, noisy/corroded/semi-faulty ultrasound could be loud as heck acoustically thru no fault of the generator, amp, or transducer.
Building a working prototype that can beam ~1W (or even 100mW) a few feet would be very interesting and would do a lot to lessen my skepticism.