Ultrasonic is not the right word. My bad. That implies an in audibly high frequency. I meant something like inaudibility low levels in the least significant bits. I think it is possible to envision that with 16-bit RBCD in a 24-bit MQA carrier. But, I do not hold out much hope for it in a 16-bit MQA carrier.
It seems to me that the Redbook compatible version of MQA probably would sound subjectively close to the 24-bit version. It was reported by 'Blumlein 88 'upthread that within a 16-bit Redbook container LSB bits 13 to 16 hold the MQA folded bands, which, of course, means that the primary band (Region A) has only 12-bits of native resolution available. However, for non-MQA Redbook playback, those 4 MQA bits would act as strong dither, improving the apparent channel resolution to below the dither noise floor.
As for those MQA bits representing the folded ultrasonic bands, 4 bits of resolution seems plenty for their purpose. Since they represent only ultrasonic content, the ultrasonic bands are not directly audible in themselves. They can, however, effectively hold the extra bandwidth needed to encode the original signal utilizing a slow roll off (time domain optimized) anti-alias filter without suffering alias products. The ultrasonic regions seem are likely captured primarily to provide wide transition band space for the slow anti-alias filter to reach it's stop band, not for the purpose of directly capturing musical information there. At least, that's my presumption.