Also, some extra info/thoughts after creating the spreadsheet and recordings:
I verified tonight that disabling L1 and L2 cache in my CMOS setup makes both Secret of Monkey Island and Prince of Persia play music correctly to the MT-32 through the HardMPU. Could this mean that the MIF-IPC-B has an unmentioned buffering and rate limiting feature?
Furthermore, I verified that with L1 and L2 cache disabled, Secret of Monkey Island plays its music beautifully through my CT2800's gameport to the MT-32. Nothing else I've tried does, but I've only tried a few things.
With only L2 cache disabled, Abuse's soundtrack seems to be correct, now. (Through the HardMPU. Haven't tried this or anything else with the MIF-IPC-B & MPU-401 after slowing down the PII.)
Tonight has been a joyous and also sad night, because now I'm understanding what's going on a lot better, but I've also concluded that I have to build even more vintage systems because of this. The reason that I had convinced myself that a Pentium II @ 350MHz was going to cover (almost) everything released from 1987 to 1997 was that in my youth, I only had Sound Blaster or SB compatible cards for my audio solutions. The Adlib music on DOS games never seems to screw up or vary at all, really, when playing them on faster or slower machines. (Neither does PCM-based music or CD-audio based music, AFAICT.) The only real problems I can remember encountering while running old DOS games on really fast machines were both graphics related, not sound related. Specifically, the hard-coded velocity of the palette rotation trick employed for water movements in Warcraft, as well as the mouse and keyboard scrolling speeds on both Warcraft and Warcraft II. Similar issues with Alien Legacy, now that I think about it. But these examples are relatively rare in my experience, regardless.
My, my, how owning vintage Roland hardware changes everything.
So in conclusion, I can put this all to rest for myself, because I have a path forward for my vintage gaming project(s). However, it looks like I've exposed a corner case for HardMPU (*especially* because SoftMPU was working correctly, even with the high CPU speed!) that should be explored and corrected. It might even turn out to shed some light on the inner workings of the whole Roland device chain. (MIF-* --> MPU-401 --> MT-32)
It's particularly interesting to me that the MIF-IPC-B prevents timing issues from screwing up the music on the MT-32 with Secret of Monkey Island while the gameport --> MIDI approach with my CT2800 does not. It's almost a logical conclusion that MIF-IPC-B is doing *some* kind of buffering or rate limiting, even if it wasn't designed to prevent problems like what I've experienced with the HardMPU. However, the fact that the SysEx delay feature of the HardMPU did not also correct it is baffling. Makes me want to understand exactly how the various delay techniques work in these different solutions.