My Quandary on Bit Depth

I should probably preface this with I am more about "practical application" in all things, but often see a massive divergence between folks who want to place their stake in the "tech spec" area versus those that "do it."
 
Will just summarize this quickly... even the "best" audio interfaces (AI) typically have a noise floor within 20 bits, and some quote the chip rather than system (so 2 bits may be "bogus" anyway). On the flip side, headroom is needed and most people use/advocate 2 bits there, so the "effective" dynamic range is about 16 bits (through the AD converter).
 
From a practical side... anyone who has connected anything to an AI to "something" immediately will be faced with the impracticality of the noise floor "spec" and find themselves working with about 60dB or so (say 11 bits for conservative application, and knock off 2 bits for headroom).
 
As far as going through an AD converter, that is about it.
 
For me, the practical application has been this:

1. Get the most range possible, and a noise print before/after each take. In many cases this is in the 9 bit range (or less). Fades get done digitally before step 3.
2. Use a noise reduction tool within SONAR (something CW really needs to consider for "everyone") that can remove that noise print - so once playing in the 32 bit realm the "noise floor" now has usefulness.
3. Finish all digital playing there, export to 24 bit, yada yada.

Even on the output side, as mp3s yield "effective bit rate of 20" (120dB dynamic range), even with all the mixing on the planet, this still exceeds the noise floor specs of most audio interfaces (and the floor is still not fully useful still due to listening environment).
 
Anyway, I just wanted to throw this out since my "stake" has always been firmly planted in the "what I can really do" side of the dune.