It's pretty good, but accuracy is rather bleh when you look at the code. What blargSNES does (and this is not to offend StapleButter, because he did a great job with it) is groups the DSP writes together as they come in, and then processes them all at once. Then the audio mixer takes the final state after those are processed, and does its own processing of the audio. This differs from how CATSFC anf SNES9X handle audio as far as I'm aware. Based on my understanding, they process audio samples alongside individual DSP writes rather than groups, much closer to how the original SNES handled it. This, however, requires more CPU power, and possible makes it harder to push to the 2nd core, which blargSNES does to relieve the main core that handles practically everything else.
You might not hear a problem in most runnable games in blargSNES because they won't make so many DSP writes, but a game like Chrono Trigger (as tested with the Satellaview music library) has music that is very active when it comes to DSP writes, insomuch that there will be missing notes from time to time.