Modern systems largely sacrifice consistent latency for more throughput and more energy efficiency. This leads to modern systems being able to do much more, but also having frequent little responsiveness hiccups and occasional larger hiccups. If you want to do a smaller job consistently, a simpler embedded system architecture is still the way to go. This is especially true when doing any kind of low-level signal generation, whether it be audio, video, or control.

As for the sound, I don't think Yamaha's implementation of "FM" [1] has been comprehensively reverse-engineered in the way that, say, the Commodore SID has. There are a lot of little quirks and edge cases to take into account when considering the full operational ranges of the various chips. Even MAME's implementation is allegedly distinguishable even after 20 years of tweaking and testing against hundreds of games, and I imagine most VSTs are using significantly less mature code.

[1] I recall reading that the actual implementation is phase modulation because directly doing FM in the digital domain would put quantization error in the frequency domain, i.e. notes would be off-pitch instead of having noise.