but what's more surprising that many folks agree with them
They hear what they expect to hear. Hearing perception is done mostly by the brain, and is influenced by many things going on in the brain, including input from other senses. The influence of expectation bias is orders of magnitude larger than the smallest real differences that are audible. That is why non-blind tests are essentially worthless. The problem is that this information is not known by the majority of the people, and this fact is exploited to the max for financial benefit by many companies.
It makes me very sad because a lot of money is being spend on things that make no to almost no real difference. And that money could have been spend on things that do make a difference: for example loudspeakers, acoustic treatment, clever room correction dsp, binaural simulation over headphones of loudspeakers in a room, ...
Yep, doesn't work for me at all, well it can almost work, until I twitch my head and then everything collapses into either a phasey mess or just a condensed centre image lacking in dynamics.
Yes, if you move your head and the spatial clues don't adapt to the movement the brain gets clues that it is being tricked and can reject the out of head illusion. Not everyone is affected equally by this though.
I think a nice comparison can be made with the following situation:
Sometimes when you sit in a train in a trainstation and the train next to yours start to move you can have the feeling that it is your train moving, and that the next train is standing still. (Or was it the other way around? Or maybe both is possible.) Until your brain picks up a clue that reveales the true situation. Then the first illusion collapses and is immediately replaced by another illusion, that corresponds with the real situation.
Another problem of standard binaural recordings is that they generally don't match your personal HRTF.
If you want real binaural then a stereo speaker system is the only valid choice.
Listening to a Smyth Realiser A16 using your personal measured PRIR and HPEQ and using headtracking (to keep the virtual speakers stationary, so properly adapting the spatial clues to the head movement) comes very, very close.