Since there's no NDA on beta 1... The new audio stack drops all hardware acceleration and is as of broken in multiprocessor mode. Internally, it seems to have been restructured, using things called audio transform modules (or something), that'll take care
of getting the data prepared for optimal output. Other than that, details are sketchy.
It shouldn't be broken in any mode, as far as we've heard, if you're hearing bugs, I'd love to hear about them.
The audio stack is a total, bottom-up rewrite, entirely in user mode. The DSP that used to be done by sysaudio is now being done with pluggable modules called Audio Processing Objects, or APOs.