I am not taking about the noise floor, I am taking about rising waveform distortion as the recording level decreases.
Certainly that does happen. That "waveform distortion" could be due to quantization errors, but it can also be due to noise or some other source. In the end all that matters is how different the resulting waveform is from the ideal.
In an analogue recording system as the signal level decreases only the noise increases, the distortion level decreases as well.
No, I don't think so. As the S/N decreases, THD will rise. Think about it - when the signal is around the same size as the noise, the THD will be much larger than one, because the power in each harmonic from noise will be as large as the power in the fundamental.
So I think what you're saying is true for any kind of recording - analogue, digital, whatever. What it means is that you need enough dynamic range so that the quiet parts (where the distortion is high) are so quiet you can't hear the distortion. So the question is whether 96dB is enough for that - and according to Meyer and Moran's results, it is.