I've tried to figure it out every time I've had to hire someone. The best luck I've had is with prior knowledge of the person (worked with them in the past, knew them from an Open Source project, etc.). I'm suspicious of the network hiring process, though, as it seems to narrow the pool too much, and certainly doesn't scale. I think it can also bring about quality degradation over time as people bring in people they like rather than people that are the best.
I just have a sneaking suspicion that some of the things that Google does wrong could possibly be attributed to hiring predominantly from the "teacher's pet" archetype.
Keep in mind this guy started in tech support. When you describe Google as requiring a "top-tier degree" you are thinking of software engineer positions. If you want to get a job at Google doing tech support, or working in a datacenter, the hiring bar is lower.
So, maybe Zeke is an exception at Microsoft, too, and this really is just some straight up astroturfing. I'm sure one could easily find a few dozen team leads at Google that don't sport a fancy degree. But, if Microsoft doesn't actually want to hire more people like Zeke, it seems like a poor choice of story to tell.
Mind you, it took him a pretty long time to do so, and lots of things could have gone wrong on the way, but its definitely possible.
Out of curiosity, how would you thin the applicant herd? Say you have 10,000 resumes sitting on your desk for a CS job. You can't read them all so you need to set a few automatable/arbitrary rules to get the pile down to 50 or so. What rules do you use? Years of experience? Google HR can't take the time to get to know 10,000 people.
What Google does is incredibly imperfect, but to trim their resume flow down to a smaller number they haven't made a terrible choice. What's a better one?