The LAME encoder I used is well the v3.97.
Be careful, in the wav files I posted, the same 2 sec. test sound is repeated 7 times, the first 2 sec is the original wav, the next 2 sec. is the 320k, next the 256, next 192, next 128 etc, with of course a total lenght of 14 sec.
So your 320k and 192k files just do sound like it sounds on the 2 to 4 sec and 6 to 8 sec segment of my file.
Regarding the file I posted for the mp3 test, you might have notived it's only the 6 last seconds that sound really crap (2 sec at 128k, 2 sec at 96k and 2 sec at 64k).
In clubs mp3 are easily spottable, because of the high volume level and because it requires more dynamics. So the test sound is a sound that exposes the algorithm to many very brutal waveform changes, in order to spot the difficulties codecs can have responding to big sudden dynamic changes. The "punchiness" of sound depends on the ability to reproduce very fast time-attack and time-decay when the waveform changes, this requires a lot of bits and tolerates hardly the compression.
Normally the mp3 codec has a "bit reservoir" that allows it to manage with such brutal changes. I takes benefit of a "hole"
in the spectrum to store them. But unfortunately I didn't let any hole in the spectrum and brutal changes in the waveform occur too often, so that "bit reservoir" is overloaded almost at once: codec don't manage with it at all at 128k, and the difference is already audible at 320k, and like you said, at 192k it is... well... "passable".
At 128 k anyway, the difference seems too obvious here to require a double-blind study... And it's a current bitrate over the web...
Le mieux est l'ennemi du bien