Higher sampling rates are not about higher frequency response, but about less aliasing though out the audible range.
Since you brought up Nyquist, yes it takes a 40 kHz carrier to reproduce a 20 kHz signal, but that signal must be precisely aligned. Think about this a 20 kHz sine carried in a 40 kHz PCM encoding, will appear the same as a triangle wave, one sample up, one sample down. If that signal shifts 90° in phase, it becomes silence. 45° cuts the amplitude in half. Only right on do you get the full energy recorded. Anti-aliasing in the down mix only serves to average the stored signal into half-amplitude also. Even a more reasonable 10 kHz signal still has pretty bad aliasing issues. But if you increase the sampling rate four fold, it's much easier to store the full energy of the signal without having to worry about exact phase alignments.
If I were recording I'd master at 384 kHz, low-pass filter at 24 kHz, and then anti-alias into a 192 kHz mix for distribution. There'd be no ultrasonic component for so-called super tweeters (nor to destabilize poorly designed Class D amps), but a very accurate representation of the original analog waveform.
