This is how I understand it.
Once Firefox e10s (Electrolysis) with sandboxing is enabled by default, every Firefox content process will be independent and restricted.
Most ALSA devices cannot handle multiple open. And the ALSA solution, dmix, requires shared memory which is a thing sandboxes do not really want to have, and dmix does not understand sandboxes, so it would probably have to be forked and modified.
So Firefox can write their own sound server to get sound data from each independent content tab, or blow huge holes in the sandboxes for ALSA dmix, or they can just use the sound server that already exists and is used by 98% of Linux desktop users: PulseAudio.