I don't know how you actually measure how long a lookup takes, but just FYI if you're interested in the number of lookups you're doing from the various caches (and in things like number of missed branch predictions you're causing), you can use OProfile or Cachegrind. OProfile is a bit more complicated to set up because the flags you need to set are processor-specific, but it causes much less slowdown.
I love performance counters and good profile tools. But the whole point of knowning this stuff is to get a handle on expected performance before you have written a line of code. Then you can make good decisions about what to do. Later the profile tools can help you fix what is broken when your code doesn't match expectations.
