A great demo of binaural sound - https://www.youtube.com/watch?v=PKdDA_jhgX8
To make this work, you would need to record the location of every sound. Easy to do for a video game (where the software knows the exact object or person emitting a sound). But it would be harder for movies, as this sounds localization data would need to be captured on set and handled during all post production work.
I did a little research and a field known as ambisonics (http://en.m.wikipedia.org/wiki/Ambisonics) generalizes this concept to more than 2 speakers. Now that I think about it I remember that the guy behind the xiph.org videos debunking digital audio myths lamenting that not enough research was being done on ambisonics.