I recall having a discussion like this one a year or so ago with Rogo, perhaps Darin, others. The solution as I recall was that the 300:1 max CR is correct, but only for a patterned type test, say a checkerboard or bright-dark lines with a high spatial frequency (i.e. small angular separation).
Real scenes are not like this, there are large areas of bright and dark, with their own edges. The idea is that transitions beyond 300:1 cannot be differentiated from those at 300:1. But, since the transitions could be say 15fL/0.05fL in one spot but say 3fL/0.01fL in another spot, we need the CR of 15fL/0.01fL or 1500:1 to not produce a hazy scene.
I have 800:1 LCD panels, recently had a 1300:1 panel, it was very obvious that I was perceiving a much higher CR with the latter. The meaning of the 300:1 is that in a 15fL/0.05fL transition one could not differentiate between that and a 15fL/0.0fL transition. Now Chris might argue against this with his felt test, but that does introduce a new edge, and you see that edge. I would maintain that yes, you do see the edge, because you are able to perceive a transition between say 0.05fL/0.01fL, as that is only 5. Now if he argues that covering a square completely with felt makes that square look blacker than another square, that would be a problem.
So my take is that both sides could be correct. We see many edges at once with our wide field vision, and all of this information goes into the CR perception. As a previous poster said, we need 300:1 at all transitions and all intensity levels to max out.