The SF2 spec implicitly expects the sample data to be one very large sample buffer, which are subsequently resampled and looped on each new note. So it's possib

The SF2 spec implicitly expects the sample data to be one very large sample buffer, which are subsequently resampled and looped on each new note. So it's possible to reuse the same set of single-cycle waveforms in various combinations and have a very small file(and I've seen a few floating around that are like this) - but a lot of playback routines do not implement this use-case correctly.