I was just born in the time of the floppy disk though, hence my concern to compile data in the memory space at true need ...:rolleyes:
But you're right, memory storage is not longer an issue nowaday

As I am running an FFUC I was thinking about saving bandwith hence further more secure the audio stream from dropouts, but if its contender is still 24 bits shaped -no matter the bytes filled in-  it does not make sense either

I maybe have a last argument to present in the direction for truncating into 20 bits direct from recording
It's well known that the last four LSBs are made of electronic floor noise, not to say simply the venue ambient and mic noises, leaving a really excellent 114 dynamic range available for music related digitizing matter when I usually set my preamps around - 6dBfs for safe margin (no limiter or compressor engaged for my classical pieces)
I experienced a pristine output while applying SRC workflow with Weiss Saracon starting from raw 16/44.1 takes and ending at 24/96 for my final mastering work, allowing the core DSP computing to apply with "pure" audio representing bits and outputting that last LSBs with either POWRx or TPDF dither that better fits from a musical sense, thanks to the "asynchronous correlation" between both
But 16bits is weak with classical music with content of very low loudness as ppp cues, that is why I am pretty much in favor of the 20/44.1 primary recording format in my case which perfectly matches with my FFUC workframe :-)

All the best