When I worked at Amazon.com they did analyze interviewing techniques, and I assume Google and other large companies do too. They didn't do controlled scientific experiments (as far as I know), but they looked at things like which phone screen questions were correlated with success in later on-site interviews, and did post-mortems of "wrong" interviewing decisions ("no" votes on someone who was eventually hired and successful, or "yes" votes on someone later found to be unsuitable). The data that comes out of this is not exactly scientific, but it does represent lessons learned from experience. [For what it's worth, a typical Amazon interview is a lot like a typical Google interview.]
Indeed. These types of interviews are much more about finding people who "think outside the box... just like me!" than about finding people who can do the job.
It would surprise me if there was any scientific evidence of this kind. This would be incredibly difficult to measure, control for other variables, arrange for a sizable sample, etc.