Another member here, who shall remain nameless for the moment, generously acquired a board and sent it along for nothing more than the cost of shipping. This should make ongoing testing much easier. For the moment, I have removed the ROM altogether, and have been experimenting exclusively with RAM banks, in the hope of answering some questions. I have some early results to report.
The most significant of those results is that the problems we have been experiencing with MIDI playback seem to have absolutely nothing to do with the ROMMaker software. Simply generating an FFF/DAT bank via GIPC, loading that and playing MIDI files produces the same output; that is to say, the stuttering drums and clicking. So, at least now I know where to direct my attention when trying to solve the problems; no amount of time spent staring at the ROMMaker source code would have led to a solution.
The problem isn't caused by GIPC alone, either. I was able to generate an FFF/DAT bank from the original 16-bit UltraSound patch set, and playback is about as good as it can get with those patches; no stuttering/echo or clicking.
Armed with that information, I decided to try another experiment: I manually hex-edited the FFF file for the 2MiB PPLT bank we have been using, indiscriminately disabling looping for almost all of the percussion. The result was no more echo or stuttering (though the clicking did remain), but playback that was still somewhat wrong.
This seems to indicate that it would be worth the time to go back to the original PPLT PAT files to carefully look at the loop points and flags, to make sure the problems aren't coming from there, and that the conversion process isn't introducing the problems.
EDIT: I forgot to mention that the echo and clicking are present during playback in native InterWave mode (as opposed to GUS compatibility mode) in pure DOS as well as under Windows. Playing one of the Doom files or GDVIB6.MID using PLAY.EXE will demonstrate this.