The point about interviewer vs. interviewee familiarity with the problem is important. It's hard to avoid comparing the interviewee's off-the-cuff solution to one that was researched at leisure and refined over multiple previous interviews. Really damn hard. Psychologists and teachers are trained in avoiding that kind of bias, and still struggle. Very few engineers are equipped to see through the "fog" and interpret a candidate's performance in a useful way. I've only encountered one - a recently former CS prof - and I've been in the business a long time.