I'm glad that both, Eric has a concern about my analysis, and you don't. Otherwise, there would be no discussion :).
With respect to the selection bias of the corpus, it may be. However, if you look in the Supporting Information, we did a control by projecting another corpus (the Alicante's 9GDB) into the factors, and we found very similar results. However, I don't exactly recall how the projection values where for the 9GDB corpus with respect to the IR principles.
You are right with respect to the Baroque overshoot. Yet, the end of the Baroque is a very arbitrary date as well, since it's the year of Bach's death. So, it could be the case that for 20 years after Bach died, similar music was still been written (maybe someone who knows more about music history can tell me if this make sense).
Anyway, thanks for your comments!!