[josephg]

There's a middleground between those extremes. There's lots of knowledge that isn't relevant most of the time, but when it is relevant its worth its weight in gold. (And often these things aren't so obvious that people will google search them.)

When I'm trying to evaluate a candidate, I want people who proactively search out knowledge in that middle zone. People who understand complexity theory, and CPU caching behaviour, and the C10k/C10M problem just because they know they might need it some day. Because they want to work on problems where this stuff is relevant.

This sort of knowledge is the difference between hearing "our website is slow - I think its the database?" and "our website is slow - I took a look and our database server has run out of open file handles. But I'm still confused - that shouldn't explain the latency we're seeing."

The CAP theorem is a great example of this sort of middleground might-be-useful knowledge. Understanding how your database will behave in the face of network partitions is exactly the sort of thing that separates good engineers from great ones.