Hi, i'm who make the spreadsheet. let me explain.

First,
It's binary rounding, not decimal. I guess that the audio softwares work in floating-point, and may also the sound hardware. So, when it convert sample to PCM integer, it need rounding and also dithering. I didn't round any sample, I used the original sample value from files I got. So, I have no clue which process the rounding take place. All I did is find the offset and compare those files.

I also make another comparison, in hex editor. You can see it here: http://twitpic.com/3ymmoy/full (see the highlight in each window)

They're bitstream from files I got, I didn't alter any of them. As you can see, there're offsets, so the files are difference when compare with foobar2000 or something like that. But when the samples were shifted to be the same offset, they're almost the same.

You can see that the difference between sample values in files occur only +/-1 or +/-2. So, I think it's must be a rounding & dithering. As I mention above, I have no clue which process did this because I didn't the one recorded these files. Maybe the player, recorder, hardware, driver or anything else...

PS: You can check my comparison with the files you have, I didn't edit them.