A lot of time could pass between recording the request count and hit
count pertaining to a particular report fetching batch, which skewed
calculations cache hit ratios.
Fix that by defering the request count recording to the end, which is
when we record the hit count.