If someone is playing the game, I will agree that he will press the directional; if he is showcasing, all bets are off. But you do raise a valid point, particularly when you factor this in:
That is more damning than the collision argument by itself.
Fair enough, I was mistaken.
The two sensor rays used for each of ground and ceiling collision while on-air are indeed vertical sensor rays (and by the way, these sensor rays aren't always cast -- it depends on the angle Sonic's speed makes); but the two sensor rays used while on terrain are only vertical for certain ranges of ground angle. Specifically, if the floor angle is between $20 and $60 or if it is $A0 and $C0, then the rays are horizontal (and they go right in the former range, or left for the latter); both cases happen in loops or steep slopes. So no, I didn't misread you but you misread me. I will concede that using the horizontal rays while walking on a mostly horizontal surface would cause the problems you describe; but this isn't relevant tot he analysis of the video because Sonic is on-air at the time, a point which you already conceded.
So you are saying that it is good programming practice to be clairvoyant and know exactly how the game will play before you have a playable tech demo of the engine? Because there is a limit to how much can be done in terms of planning ahead -- games are very complex systems, and the effects of one small decision can have many unintended consequences that you can only find out by trying it out and seeing. Moreover:
I hardly consider tweaking the collision routines to be 'rewriting the base engine'; particularly given that they already have collision routines for checking collision along multiple horizontal rays. Instead, I would consider it a tweak. I will see if I find the documentary you mention.
Any such quantum mechanical effects would be drowned out by decoherence and thermodynamic effects, so we can safely rule them out.