It remains weird that hd is faster than hs under certain conditions. I have checked that it is not due to a transcription error, at the least. It is the case on both rise and fall time. Do note that this does not occur for the parasitic-less netlists, only for the PEX extracted netlists.

If I had to take a stab at the answer, I'd guess that the issue is that HS cells are not high density, so for example the most basic

inv_1

cell is 1.38um x 2.72um in HD but 1.44um x 3.33um in HS. Given the very high resistance of local interconnect, that dominates the parasitic extraction and might be a big hit on the performance of low density layouts. Just a possibility.