The application has been configured to update the cache before every read, which doubles the work performed on each request.

The cache uses too much memory on the application servers and is being flushed to disk before it can serve any requests.

The cache hit rate is too low, so most requests miss the cache and continue through to the database with little load reduction.

The database is faster than the cache for the data being requested, which makes the cache layer counterproductive in this case.