Do you mean reverse the polarity of one of the channels? I would imagine that sounds weird and confusing, and testing it bears that out IMO.
What I've found is that you can get quite good stereo imaging if, rather than just changing the relative volume of the channels, you correct for the different times it takes a sound to reach your ears as well. Our brains can detect even quite a small difference in the Doppler shift heard by each ear, and it takes that into account when trying to place sounds.
For comparison, here's a weird noise with three different panning algorithms: the first is a naive algorithm that multiplies each channel by amplitudes whose squares sum to 1. The second is based on multiplying each channel by the inverse of the distance from the corresponding ear, with no Doppler shift, and the third is the same thing but with Doppler shift. I think the third one sounds reasonably convincing.
e: Note that none of these algorithms is able to distinguish a sound that's in front of the listener from one that's behind. That would presumably involve some complicated filtering and I wouldn't know where to start.