The fact that the Atari STs had MIDI built in was the main reason that they were so successful as computers in recording studios (and that they were much cheaper than Macs certainly helped a bit as well)
Yeah they were kind of known for that, although I personally think this was also a bit of a myth.
Even back then there were add-on interfaces like the Steinberg Midex which used some kind of timestamped buffered timing, which was probably even better. And many synths actually had quite bad latency anyway.
But for sure, having direct access to hardware and a non-multitasking OS made things much more deterministic "by default" and that was certainly an advantage for this use case.
That Atari didn't do anything special. The MIDI went in/out of a standard interface and the interrupt was - IIRC - on the keyboard line. Midex was a step up, because the messages were pre-buffered, which was a little quicker. But MIDI timing was pretty bad anyway, because the serial line was too slow.
It was tolerable if you put every synth/drum machine on its own separate buffered line. But when you had four or five devices connected to a single MIDI out on the Atari - ha ha ha no.
The absolute best timing resolution was around 700us, which is fine for single notes - although that didn't allow time for the steam age 8-bit processor in the target synth to wake up and notice it was supposed to do something.
But when you had a four note chord, it never sounded truly tight. And if you added some aftertouch or mod wheel - oh dear.
So you ended up with this weird cultural thing of music that was quantised and on-the-grid but also had quite a bit of timing slop.
Big-name bands often used synchronisers and tracked each synth line to tape in a separate pass.
USB over MIDI is much better because there's basically zero latency and everything (usually) works as it should.
