I gotta ask... what is the point of creating a module file that's larger than an MP3? Just to see the fancy display of a tracker program as it plays?
Is the CPU load of playing such a complex module file really any lower than playing an MP3 file?
That's an excellent question. With a 32-instrument, 15MB file .xm, the composer wasn't going for size or efficiency, but rather was just making music using a tool he was familiar with. mp3s were not common when he distributed that song, so he just distributed the .xm module. He is still active, and his stuff is mp3s on bandcamp just like everyone else
rather than the ableton/protools/reason/etc. project files that created them.
IIRC, it takes a 486-100 with a special optimized DOS player to play back MP3s at 44KHz, and you're right about a 486-66 being able to play back at 22KHz. The deficiencies are because of the floating-point speed, which had increased dramatically in the Pentium. Any Pentium at any period-accurate speed should be able to play any mp3 just fine.
I have no idea if a 486-66 is enough power to mix 32 16-bit tracks @ 44.1KHz realtime, but that would be very easy for someone to verify, and cool to know. Were someone to perform this test, I'd load up the song, set the playback rate to 11Khz, start playing, seek to the section that has the most channels going at once, then start bumping up the sample rate (you can do this while it is playing in Fasttracker 2) and see what samplerate it croaks at. If you get up to 44.1KHz, start upping the mixing quality (but once you hit "sinc" it may crap out because sinc interpolation uses floating point IIRC).