I tried this again with 3.3.2, Got the same results (file size is the same for 16 and 24 bit saves). I save a recording I made by exporting an audio file which brings up Flac which I wanted, than saved each of the files. I also tried saving as a ,wav file to see any differences and they were what you would expect. The 24bit ,wav file was much larger than the 16 bit ,wav file. This problem seem to be with saving as Fflac

I ran the Amplify effect at -1dB to slightly change the volume and “fill-in” all of those zero bytes with rounded values.

24-bit FLAC = 33.5KB

I re-opened the original 16-bit file, exported as MP3 (again to create lots of rounding) and then I re-imported the MP3.

24-bit FLAC = 33.2KB.

P.S.
A bit more about “rounding”, etc. If you have a 16-bit sample you can convert it to 24-bits by left-shifting 8-bits and filling the 8 new least significant bits with zeros. You get a bigger number but there’s no more real information and those useless extra zeros are easy to compress.

If you do any processing on the 24-bit file you’ll get different numbers in place of those zeros and it’s not so easy to compress.

Floating point in binary is more “complicated”. but the same kind of thing happens.

A couple of imperfect analogies… If you have $100 in the bank and divide it 3-ways each person gets $33.33 and now you have to keep track of the pennies so that takes 2 more digits to the right of the decimal point (and somebody gets an extra penny).

Or, let’s say I have a nice whole number (integer) like 1000. Or we can add decimal places and write 1000.0000, etc. If I put that in my spreadsheet and reduce it by 1dB I get 891.2509. (I could show more decimal places and I don’t really know if it ever “ends”.)