It’s possible their compression settings were actually exaggerating the peaks instead of compressing them, and then they did nothing to control those peaks afterwards. This is a really common thing that can happen with a superficial use of compressors. Especially if you did averaged loudness-matching of the compressed signal with the uncompressed signal. It ends up being spikier than before compression. I would entirely believe a waveform with those added spikes would be more damaging than a controlled waveform that had been saturated or limited after compression. I don’t have access to the original publication, so I can’t check and find out.