It's not about compression. There's BRR compression on the SNES, but 99% of the time that's not what people are going to mean when they try to describe the way the music sounds. The main form of sample optimization on the SNES is to just physically cut down the length of the samples and then assign loop points to simulate a sustained note. Usually to the point where an instrument just regresses into a simplistic waveform oscillation after the transient.
Oh that's fascinating actually, and makes a lot of sense.