I think there are 3 great mysteries in the modern age:
1. The nature of dark matter, if that is what it really is
2. The nature of dark energy, if that is what it really is
3. The nature of autofocus error reports: equipment failure or user failure?
As you point out, the issue of giving a focus point (or a cluster of points) two targets at different distances is also a very complex issue. My D2h would consistently and repeatedly pick the background target, which was the biggest problem I had with that camera.
The problem was compounded because the sensors were very widely spaced and it was well known and easily proven that the actual AF sensor extended far beyond the small viewfinder reticle outlines. If you gave it one unambiguous target it was absolutely state of the art.
My D200 improved on that particular issue, although it was generally not in the same league in terms of tracking fast moving targets, and I found that the D300/700 improved things further, but it may also have been simply the benefit of more and smaller AF sensors. And with those cameras (300/700) I never found that the AF sensors extended very far, if at all, from the reticle outlines.
And the 51 pt AF viewfinder reticle positions are packed so tightly that if it did extend beyond the reticle then it would arguably be "sensor bleed" where the adjacent sensor was used. In other words... is it possible for two adjacent sensors to be reading the same area, and if not then it would be impossible for the sensor to extend beyond the indicated reticle outline.
I think you are taking the correct approach- methodically studying the results and not assuming that any deviation from your expectations are some sort of "failure".
The most interesting aspect of the D600 (to me) is the new f/8 specification. It implies that even at wider apertures the AF engine should be more responsive to difficult targets- dark targets and low contrast targets. But that is just a hunch.