That MT32 has about a dozen great sound in it that haven't been duplicated by more modern forms of synthesis. The MT uses "Linear Arithmetic" synthesis where sampled attacks are used over synthesized sounds.
The old FM synthesizers like the TX81z and the DX7 make some melodic percussion sounds like vibes and some electric piano sounds that also hold up to the test of time.
My VL70m which uses "Physical Modeling" synthesis can emulate a saxophone better than any other sound module I've ever tried, and I am a sax player.
The secret to recreating acoustic and acoustic/electric instrument sounds in MIDI isn't as much about the tone as it recapturing the nuances that those instruments allow the player to exploit.
The trick is to listen intensely, not to the tone, but to the notes and all all their variations. You will hear sax players often scoop up to pitch, variable vibrato speed and intensities often on the same note, which way the vibrato varies from center pitch (mostly under for sax lip vibrato and mostly over for guitar finger vibrato), the ornaments the instruments use and so on.
Different forms of synthesis are better at emulating different instruments.
Here is a synthesized sax clip. The tone is thin because it was recorded in 2004 with a pre-iPod Archos Juke Box with its internal mic hung out near the PA speaker on the gig. clip
It was done with a Yamaha WX5 wind MIDI controller and a Yamaha VL70m synthesizer. Even with the think tone it sounds more like a sax than any other type of synthesis. Everything but the vocals is played on synths.
Here's one I did with the VL70m in 2008 emulating a lead guitar. This was also done on the Archos so the sound is tinny. As above everything but the vocals is synthesized clip
It's more about expression than tone. After all Jimmy Page sounds good on his LP, a Tele and even a Danelectro. Hendrix sounded great on his Epiphone too.
By having a rack full of FM, LA, ROM based, and other forms of synthesis, all with a latency of about 6 or so ms, I can mix and match, taking the most appropriate sounds from each synth and mixing them without having to time shift any tracks to compensate for different amounts of latency on different software synths, and with no load on the computer CPU to make the sounds so I can even record at a higher bit rate if I want to.
Someday computers will be fast enough to match the speed of the old hardware modules, and that's when I'll switch.
Notes