[tdavis]

I think this bottleneck is easily (and still amazingly) noticed when storing things in RAM rather than on disk. For instance, immediately after a restart of our production database it can take up to a minute to run certain queries. Give it a few seconds to load everything into RAM, though, and it's nearly instant. I've noticed the same thing when switching from file-based to RAM-based queuing.

Everybody thinks, "duh, RAM (or SSD or w/e) is faster than disk" but sometimes these sorts of real-world tests still amaze.