I was curious why mp3Gain always seemed to report clipping, while I could hear no distortion (even with files 12 dB or more over the max no clip level). After playing with the gain and analyzing the decoded wave files with Wavelab and Soundforge (which were not detecting any clips), I realized that mp3Gain was reporting clipping when as few as a single sample was over full scale.
When such a "clip" is decoded into a wave file upon playback, there is NO clipping (i.e., flat-line at full scale). It's irrelevant for all practical purposes. It is an overshoot by the encoder 99.9% of the time.
I gave up my search for a true clip reported by mp3Gain (at ~89-92 SPL) after analyzing more decoded wave files than I care to remember.
Why not define clipping as at least 3 consecutive samples over full scale as professionals do -- and this is conservative (see http://www.digido.com/bob-katz/level-practices-part-1.htm\)l)? This leads to a true flat-lined full scale clip, which may be be audible.
As it is, I've found mp3Gain's clip detection to be meaningless the vast majority of times at normal volume levels.