Totally agree that the aim is to minimize false positives. although I think you could devise the pair programming type interview to still allow for 1 hour chunks with many different people doing the sessions. A good phone screen should be able to filter out candidates who obviously don't have the foundation, and watching them code in a (semi)realistic environment can tease out things like resourcefulness, search IQ, etc. I find those things to be better indicators than the ability to find the distance between two nodes in a binary tree.