Whilst explaining verbally to others your every single step of your thought processes while you are doing fine-grained analysis the problem (something which one almost never has to do in real engineering environments), and under incredibly unrealistic time pressure ("these people will either love me or hate me, depending on what I say in the next 5 seconds.") And typically to solve "problems" that bear little or no resemblance to the kind of problems that you'd actually be tasked with solving on a day-to-day basis at most of these roles. But which instead seem to solve other, ulterior purposes -- in particular, to shore up the interviewer's own insecurities:
"Gosh, I wish I knew how to find good people for this job. And heaven forbid if the candidate should ask me to explain our business model... But if we ask a bunch of Google-style questions at the whiteboard... I guess that means we must be in same league as Google."
Fine-grained analysis? Again, you're not expected to be a master, but it's not asking a lot for you to break down a seemingly complex problem into something workable (e.g. The maze solver example above could be approached with choosing a right turn whenever you approach an intersection...does that seem very fine-grained to you?).
Incredibly unrealistic time pressure? Again, in most interviews I've had/conducted, you're not expected to just know the answer. If you do, that's great, but you make it seem like they're waiting to pounce on you the moment you struggle. I've never seen that in interviews for larger companies.
That being said, everyone knows you need to have a solid understanding of data structures and algorithms to be seriously considered, but I don't think Google is necessarily looking for software prodigies.