For most applications the optimal size of a (time-based) cache is around the point that keeping an item around in the cache is equal to the cost of accessing it again… A side effect of that is that systems that run faster, and therefore access each cached it more often, get more benefit out of a cache by pushing that set point out further… The bottom line is that both locality and memoization are huge contributor to the performance of systems, both positive and negative. - HN

see also

• Exploring How Cache Memory Really Works