I seem to remember in this era (I'm a huge Super Socket 7 fan) that even the mostly godly disk was maybe putting out 70mb/s (unit?) so most controllers, even if they did support Maxtor's ATA133 it didn't really matter outside of synthetics. There was no "shame" in only being able to run ATA100 because it simply wasn't the bottleneck. In fact many times if you were running overclock bus speeds like 75(112) or 83(124) a few final straws to try and stabilize that OC would be to step down to ATA66 or 33 - though if you're down to 33 it's YOLO time basically.
I guess my tl;dr is that most southbridge/EIDE drivers of the era prioritized doing anything to lower random access speed vs. max throughput, since nobody dreamed we'd have a single drive that could saturate an equivalent ATA600 link! I think it's pretty amazing people have gotten 90mb/s cause there's no way they could have tested that back in the day with an actual drive - it would have to have been some synthetic test rig blasting bit patterns or something.
So I wonder if setting lower ATA speeds of 66 or 100 might help these adapters? Since really the ATA133 spec really was a post-spec optimized for the Maxtor drives of the day? (I don't mean that as a negative, I give Maxtor a thumbs up for that!)
Sorry for the ramble, but keep posting your results - probing the edges of performance on these old systems is super fascinating to me.
Off Topic:
I'd love to see sometime tests on if things like the special inter-chipset interconnects used by say the SiS chipsets and such actually made a difference by bypassing PCI bus. Or even on older PCI driven southbridges if IDE speeds really were influenced by bus overclocking.