Most of you guys are overthinking this. You don't need to re-encode anything. In the game, all the sounds are mono sound sources placed about the player in 3d space, and then the game figures out the stereo representation based on the player's position. All that is needed is to improve the algorithm so that it takes the position of the head into account as well. What Veqtor described would be an ideal implementation, but we can begin by just placing the virtual microphones on a virtual head and linking it to the headtracking.
Wouldn't it already be linked to headtracking? I mean, if you play Half Life 2 and you move your mouse around when someones talking, you'll hear it from your left side or your right side or wherever the source is relative to where you're looking. Isn't moving your headaround with OR the same as moving your mouse around? I haven't been able to test this out on the OR yet.
I think to get a quick and dirty immersion effect you'll need to have a virtual microphone for each ear (which will shift along a 2D plane) and then perfect the reverb of the 3D environments based on the players position in the room.