Thanks for the explanation, make more sense now

What annoyed me is the reason they gave...

"A core2 duo can't decompress audio and play the game at the same time", that's complete bullshit. My old computer can play mp3s, play a video or two on you tube and play a game just fine without any noticable effect on pretty much any game... If it lag, it's because im low on ram or something, not because of the game sounds...

They never specified the audio codec they were using, only that it required software decompression.

It was almost certainly not MP3.

Quite likely they were using the XMA2 format since it is hardware supported on the platforms and offers very tight compression with multi-channel sources. Based on the various homebrew forums that have reverse engineered the format, it looks like decoding it is very compute intensive but results in very high quality audio.

makes sense.

but yeah, probably for reasonable real-time audio decoding/playback, a format which allows a low-complexity decoder makes sense.

It's impossible to call them out on that without having more data.

How many compressed sounds are they streaming/playing at once? 100? 1000?
What kind of throughput can your PC get when unzipping files? 100MB/s? What if instead of dedicating 100% CPU time to that, you can only spend 3% (that's ~1ms per frame for a 30Hz game) because the other game systems need the other 97%?

Now why ZIP? Because rumour has it that the Xbone has dedicated hardware that can perform LZ decompression "for free".

BIG EDIT:

at first I partly disagreed, thinking that 100MB/sec seemed pretty low.

I was wrong, partly. running some benchmarks it seems your prediction wasn't too far off for "typical" data. for a test with a 10MB DLL I am getting ~162 MB/sec (for a single thread), so it seems I was a bit overestimating its speed.

more testing. seems to depend somewhat on the specific file tested and the specific decoding logic used (for example, 205 MB/s with a big glob of C source, and 125 MB/s with a different DLL), so yes, if one allows for weaker hardware and/or a not particularly optimized Deflate decoder, then 100 MB/sec would appear to be a fairly reasonable estimate. (I had just expected it to be a bit faster, apparently a bad case of mental "Dragon Ball Z" numbers going on...).