Since I’ve done a bit of binaural recording, let me chime in with a couple of observations.
Regarding pinpointing sound sources - humans are really bad at that in a wide zone straight in front of us, extending over our heads backwards. The reason for that is simple, we are primarily visual creatures. We are better at locating sounds in the perifery of our vision and outside it, enabling us to turn our head/eyes towards the sound(threat) and locate and identify the source. That’s why we don’t need sound to be emitted from within a screen, if we see someones mouth moving on screen, that visual cue will dominate over sound of the voice being emitted from a somewhat different location, and the mind will fix the discrepancy for us.
Headphones will always be superior to loudspeakers at recreating a sound space, since the sound of loudspeakers inevitably will reflect (and at certain frequencies, resonate) against the boundaries of the room, superimposing on the sound space recorded/created.
Binaural sound does not have to be recorded at the ear drum, that just causes a lot of problems. (It is however our research heritage, from studies of auditory function). The reason is simple - if you reproduce a sound recorded in the ear canals of a dummy head using headphones, the sound will pass through two outer ears . Those of the dummy head, and those of the listener. You can try to reverse the process after recording if you are hell bent on using your dummy head, but that is just convoluted and creates problems. You should simply record just outside the entrance to the ears, and all the important cues from shoulder/chest reflections, the frequency dependent head occlusIon, the phase difference... will be preserved. And the sound will only be coloured by the ears of the individual listener, which is quite important since ears differ a fair bit.
Obligatory names-dropping - I almost bought an Aachen dummy head off Ken Kantor of NHT speaker fame. I wish I had, it would have been an awesome conversation piece, but I was at a financial ebb, and I had no real use for it. This was before digital processing had really taken off, so unless you used it for measurements, actual recordings through it were flawed as per above.
I think it’s great if sound space simulation takes off, we can potentially do so much today! Can’t wait to experience good examples in future games. But I’m bad at discussing these things. I used to really care.