Originally posted by slimdave
Hit ratio is a waste of times.
In this case I don't agree - if the hit-ratio comes from a representative sample of activity and you monitor it as you reduce the buffer size, it should give an indication of when you've gone too far. End-user response time is of course a better statistic, but harder to measure.

(Hit-ratio is also a good indication of your success on "International Smack a Developer Day").