mp3tag stores audio analysis and undo information in tags (though it uses APEv2 instead of ID3v2, just to be annoying). It also stores replaingain information in the Xing header or LAME Info header (which are virtually the same thing). It also seems to modify each mp3 frame to change its gain setting.

It seems that the latter part of that ought to affect playback on the player. I don't know why it wouldn't, really. Perhaps the tracks you chose were already close to the ideal that mp3gain tries to match. Perhaps the player somehow undid itself by reading both the gain info in the frames and in the tags (this seems highly unlikely).

See if you can find some real quiet track and a real loud track and mp3gain them and play them on the player. Let us know if they seem closer in volume or not.