Unless I understand compression all wrong, isn't the point to make the quiet parts loud? So in one case you have a song peaking at 102dB and in the other case virtually the entire song is that loud.
It's already known that hearing damage is cumulative, longer exposure to loudness is bad, so it seems like a common sense result in line with all existing research.
The pre-compressed and compressed versions were loudness matched to be the same overall loudness, according to the description. My guess is that they set the compressor to actually make the waveform spikier, without fully understanding what’s going on. Just a guess, though. I can’t check to find out.