.Several months back, I did an informal test myself with different-encoded mp3 files via whatever LAME version was out then. The test music sample was a minute of classical orchestral music from "Jupiter" of The Planets (Dutoit, Montreal), which included part of a tutti section, some exposed woodwind solos, crash cymbals, etc. I don't even listen to classical music that much, but I figured that that this would be a worst-case scenario for audio compression.
I created samples at CBR 128 kbps, CBR 192 kbps, VBR -V0, and CBR 320 kbps from the CD. I didn't know about foobar's ABX plugin, so I just had a playlist with all the lossy files plus the original with a randomized playlist order and hit "next" via a hotkey 10 times between listens so would have no idea which file I was listening to. With this method, I might get the same sample twice in a row. After making a guess at which version I was listening to, I looked at foobar to see which it was.
After a lot of trials, I was able to tell apart the CBR 128 kbps and CBR 192 kbps from the rest with high accuracy. Mostly, it was the string section entrances and woodwind solos that were the giveaways for the most lossy samples. Maybe I should have tested VBR -V2, but the -V0, 320 kbps, and lossless sounded the same to me. If I guessed between those correctly, it wasn't by a significant margin at all.
For other music, I'm sure I could get away with a lower bitrate than VBR -V0. I'd heard a lot that formats other than mp3s had better quality at many bitrates, but I didn't know which encoders could rival modern LAME mp3 encoding. Maybe I'll try Vorbis instead next time. aoTuV seems promising to me--thanks for the info.