I was curious about this topic so after watching a few convoluted videos I came across
this video from MIT Open Course, which I found very interesting. If I understood it correctly, they are saying that on the transverse plane, we mostly use time difference to localize sound for low frequencies (most sensitive at 750hz), and sound level difference for higher frequencies (maximizing for 5-6K range). Spectral difference plays a role as well but they made it sound like it affects localization more on the saggital plane than on the transversal.
If you go to a sound generator, set it to 6K and increase or decrease 6K for one ear via EQ, you can actually hear the sound "rotating". Effect is noticeable even as low as 1db. It also works well for 750hz. I also tried 250hz and 4K to test some of the theories they mention in the class and indeed, you get some effect but it is much less pronounced, and requires much steeper adjustments so that was interesting too.
After some experimentation, I think I was able to introduce/improve instrument separation on flat sounding headphones a bit. With Take Five for example, you can make it so that the sax moves in one direction and the bass in the other, creating somewhat of a sound stage on 6xx which mostly sounds very in-your-head to me normally. It was minor, did not work as well for every song, and for vocal songs " locked" the singer to one side of my head mostly so I'd not call it a breakthrough just yet but It is definetely fun to experiment.