Oh right, I read that wrong. Thought you did per-voice 3D positional encoding on server but mixing on client.

Interesting use-case, got my brain racing off to find silly solutions.