I proposed to my manager (although this proposal was rejected) that we try something like putting our high performers through our interview loops. This would let us calibrate our questions and answers. If you're right, that intermittently otherwise good candidates flub simple questions, I feel like that would come up in the test interviews - and maybe there could be a good compromise that somehow takes this into account.
Unfortunately, I think that whilst it's a good idea to run a few mock interviews with "known good" people who already work well within the company, I also think that you can't replicate the stress that a potential candidate might feel where the stakes are non-existent, and a rapport already exists between interviewer and interviewee.
A good rule of thumb I've heard is that for an hour interview, you should expect people that work for you to solve and complete it in about 20 minutes. If it takes them longer than that, the question will be too difficult for candidates once stress is factored in.
