When comparing the softsynth to a hardware solution the hardware solution will, possibly, sound better (it's mainly the reverb and antialiasing algos that make the difference.) However, when comparing two GM/GS solutions, be they software or hardware, to the MT-32, they both sound equally bad.
These being said, I had a SCC-1 in my old Pentium, and the difference between its output and the WIn2k soft-synth was actually not that noticeable to me - keep in mind, though, that we're talking '90s game music here, simple, nothing too complex on polyphony and multitonality. The only difference I could notice was the reverb and/or antialiasing I mentioned (the SCC had a warmer and deeper quality).
Cheers,
V.