I´m not sure what I´m about to write, but I think it goes like this:
If he wants to clip the peaks until, for example, 50% of the threshold knob, instead of overloading just one GClip with threshold on 50%, he puts one GClip clipping on 55% (cutting half of the peaks he wants) and a second on 50% (cutting the rest of the peaks). This way the "final GClip" (in this case, the second one) won´t have to "deal with too much peak data" (which could lead to a nasty distortion), as the first Gclip will be already cutting part of the work. Besides threshold, all other settings on GClips would be the same (proably Gain 0, Softness 0 and Oversample ON).
At least this is the same logic that I use for putting a GClip before the SSL Comp. I cut the highest peaks (probably snare hits) with GClip so the mix won´t punch the SSL Comp too hard, and I can set a lower threshold on it to reach the overall -4dB reduction that I like.
If I don´t put this GClip before the SSL Comp (or on the snare track) I will probably have to make SSL work much harder (compressing much more than -4dB) in order to control the snare peaks AND compress the whole mix. And it will probably ends pumping the whole thing and/or destroying the snare tone. By using the "GClip -> SSL Comp -> Gclip" chain before the L2 I can make a very "squared look" mix with very little dynamics (at least visually), so the L2 won´t have to squash too much the peaks (which would cause nasty digital distortion) as I aim for the desired RMS.
But I can be wrong! : )