The usefulness of having a discussion on why a 196 dB noise floor is better than a 144 dB noise floor while listening to a file with a 96 dB noise floor on a reproduction system with an 80 dB noise floor is limited unless we consider processing, truncation and the effect of how digital audio can collapse if it's allowed to truncate.
The accumulation of truncation artifacts, even well below the thermal limit can have an effect with certain types of source material like the depth and spatial perception, ambience, punch, separation of instruments, panning width and realism get compromised. Dull, boxy, mushy digital audio. It's not about being able to hear noise or distortion at subsonic levels in a blind listening test. More like what the fuck happened to the reverb tail on the snare?
Increasing dynamic resolution should (and does) make truncation artifacts smaller. Dither prevents those artifacts from remaining in the audio, and it needs to be applied any time the audio gets quantized. The output of your converters, the output of plugins and the output of the mix engine. Throwing more resolution at it makes truncation less of an issue when processing, but if left unchecked, there's no guarantee that it won't bite you in the ass later.
But more is more. Maybe it's a marketing thing.
More of an audio thing would be to design a digital system that runs at whatever resolution you want, and carefully counteracts truncation properly at every stage necessary. I think this is the line of reasoning behind Harrison Mixbus.
The next step would be to have a reproduction system more like a mid to high end stereo from 30 or 40 years ago, capable of better reproduction than Youtube or an iPhone. Not sure how much people actually care about that these days.