Cache usage is unusually low in the gc logging.
This looks likely that it is due to a problem in the expiry date calculations, leading to items being garbage collected early.
This could cause a performance problem in terms of more round-trips to the server.
start the agent, open any protected resource, login, check agent log - entry is validated against policy service and stored in cache.
wait a little (5 sec), refresh page, check log - entry is read from the cache
wait 20 seconds, refresh page
Resource is retrieved without a round-trip to the server
refresh page -> agent goes out to AM (policy call) instead of reading it from cache