In a previous life (or maybe it just seems that way) I did some R/D work using small directional/noise canceling mic’s to help cancel out distant audio while picking up near sources. Basically the mic has equal openings in front and back (the back normally has two small holes to accommodate the need for the electrical connections). The theory is that the more distant, or off access, the sound, the more likely it is that the audio will enter both sides of the mic at the same time and cancel out, while nearby, or on axis, sounds tend to enter one end sooner than the other and therefore does not cancel out. I would also agree with AWOL that a “cross”, or just dual, configuration would probably be more effective (greater delta between sensors, like human ears) than using more sensors, especially if each sensor was acoustically isolated from the others (think sound absorbing foam). Some things that could make things easier would be if the sound was in a short duration pulse source (as previously mentioned) i.e. a clap. Also, higher frequencies might make direction detection easier (long waves like sound can be difficult to work with over short distances) so an ultrasonic source could make things much easier (not to mention undetectable by people).
This sounds like an interesting project. I hope this is helpful and that you post your experiences.