Had too much spare time today (snowed in once again), so I wrote a small script that detects hard-clipping in wave files. Ran it over DT12, and probably not too surprising, it found a ton of clips. This one is from TBP, right before "would you talk me off the ledge":
That's a solid 10 consecutive samples on the right channel hard-clipped.
What's interesting is, they are not clipped by the maximum a 16-bit waveform can handle (i.e. the value 32767). They are hard-clipping somewhere around 32,000 but still have a tiny amount of wiggle there. I think what that means is that it was instead the compressor component that was driven to the limit.
EDIT: Here's another interesting graph. It's the distribution of amplitude values over the whole album, zoomed in on the tail end of the distribution (towards the 32,767 value). It's coming in normally from the left as you would expect from an audio signal, but then you see those two big spikes.
The first, smooth spike will be the compressor one, whereas the second one is the 32,767 one, i.e. where they actually maxed out the 16-bit waveform range.
The way to look at these spikes is, all that "mass" of those two spikes was, in the original audio signal, to the right of those spikes. Meaning, a solid chunk of dynamics got smooth-squeezed by the compressor, and hard-clipped by the waveform range. Not good.