Great writeup. But isn't this only true if you upsample? A 16 bit stream otherwise played on a DAC capable of much more will be still 16 bits, just padded with a bunch of 0s, no?
From my understanding, upsampling has nothing to do bit depth resolution in the context explained above. What I said about digital attenuation is true whether we are talking about 44.1K, 48K, ... or 384K.
Meanwhile it is true only when there is no upsampling; indeed, as soon as you start upsampling, you are creating a new file and presenting something else to the DAC chip. Some DACs behave differently whether they are used at 44.1K or 192K (the Benchmark DAC upsamples all incoming streams to 110KHz because their DAC chip performs best at that particular frequency according to its designers).
So I should have probably been more specific, by saying that we are attenuating for instance a 16/44 stream using a 24/44 calculation.
Regarding what you said about a 16 bit stream played by a 16+ bit DAC, I do agree with what you said to some extent. The 16 bit data won't be magically transformed to something else just because it is played by a 24 or 32 bit DAC (i.e there will be no creation of new data).
However, by looking at how dac chips behave, presenting a 16 bit stream padded to 24 bits can in some cases provide with better results at the analog outputs. Let's us take the PCM1704 dac chip for instace: if you look at page 4 of its data sheet (see here: http://www.ti.com/lit/ds/symlink/pcm1704.pdf), you can see that the same -110 db signal looks like a sine wave when played at 24 bits but is barely recognizable when played at 20 bits. Given that the -110 db is above the noise floor of both 20 bit and 24 signals, it shouldn't matter if the DAC behaved similarly whether it was fed with 16, 20 or 24 bit data.
As a result, padding with a bunch of 0s and presenting a 24 bit stream (of 16 bit file) to a 24+ bit DAC might change or improve the sound under some specific conditions.
I hope this helps answering your question.
