TL;DR, but skimming through an interesting problem comes to mind: The OP "orthogonalizes" the question of the sample accuracy's (24 vs 16 bit) and sample rate's (48 vs 192 kHz) impact on quality, answering one independent of the other. But even with my limited background in mathematics it's quite obvious that that approach is not entirely correct: the Nyqvist theorem only really applies when you have infinite sample accuracy. It would be interesting to see an analysis of how the two interact; i.e. how the discretization error impacts the highest representable frequency.
The highest representable frequency really is the Nyquist rate: you only need 1 bit of sample depth to generate a digital signal whose corresponding band-limited continuous signal is a sine wave that oscillates at that rate.
Sample depth roughly tells you how far away a sample can be before the magnitude of the ideal sinc function used to reconstruct the continuous signal falls below your quantization threshold. That distance in turn gives you an idea of the frequency resolution you have... You could in theory run an FFT over that many samples and detect that the corresponding change was not just due to quantization noise.
