Microcaching is fantastically useful. Way back when, when Edgeio hosted Crunchboard (the Techcrunch job board), we added it to deal with the load it imposed. I don't remember if I'd read about it anywhere first, or was just desperately trying things to avoid it hammering our database.

[For anyone not clear on what is meant: Microcaching is the idea of doing caching with really short time intervals, maybe as low as a second. The point is to set the TTL for the cached data so short that users will rarely if ever notice stale data even for dynamically updated content. The upside is that even a 1 second cache time for a heavily requested page means only 1 request per second per frontend for that resources hits your backend, no matter how much your traffic increases. Super-short TTLs often allow you to cache resources that you otherwise couldn't, and still get a hard upper bound on the load they'll impose]