Hmm... well it's interesting.

Though I think that you'd have to push OpenAL to it's max to make it work properly...

Heavy rain might be hard to do...

Though... just had a :!: moment...

What if instead of relying on OpenAL to do all the mixing, you had another layer which separated these 'sound emitters' into slices of pie around the listener...

Lets say, oh 32 slices. (or less if you want to have music playing as well.) You'd then only have to software mix those sounds together that fit into a single slice of pie segment. Pump those into OpenAL and let the API and hardware take over...

Trouble now would be creating a demo.