You could maybe compare some performance measure (response time, or whatever you'd expect to change as an effect of the caching) for repeated tests with the same input, under otherwise identical conditions.

This could in theory allow you to test (1) whether caching is being used at all, and (2) if you're getting the same results in case the caching is being used.  (But it would assume that you can start with an empty cache...)

Just some thoughts...  What exactly are the tests supposed to prove?