I have been emulating many different systems (including SNES) on lots of platforms, and have almost never been bothered by lag in the emulation itself.

But one day someone was harping on about lag in an SNES game, so I loaded it up (Android RA SNES9x core) just to double-check before arguing with them... and indeed found the lag quite noticeable. But I also found two ways to reduce the lag significantly.

The first thing is to try different cores. There are multiple versions of SNES9x available on most versions of Retroarch, and there's a reason for it. The different cores can give surprisingly different results for a single game.

The second thing (which had the greater impact for me) is to enable Retroarch's look-ahead emulation for one or two frames. Just as it sounds, this will cause the core to emulate the next n frames with every possible combination of inputs. It increases the processing requirements exponentially, but for something like the SNES, many platforms can handle it. I know it seems counterintuitive (if it can emulate hundreds or thousands of possible frames in realtime, why can't it do just one?) but it worked for me. It must be some kind of throughput vs. latency thing.