Description:
------------
In our environment we have a problem with APC. It's 
described on wiki as thundering herd problem -
http://en.wikipedia.org/wiki/Thundering_herd_problem

If we have some long queries that are cached in APC and 
5000+ users using website. When cache expires ~300 users 
produce same query to DB that stalls it for some time.

You can use RED ( 
http://en.wikipedia.org/wiki/Random_early_drop ) to avoid 
this problem.

For example if TTL for cache record is almost ended (e.g. 
((creation_time+ttl)-current_time)/ttl < some_tunable) you 
can fail apc_fectch() for this record based on probabilistic 
function inverse proportional to ((creation_time+ttl)-
current_time)/ttl.

For now we workaround it using cronjobs or emulating RED in 
our php code, but with time it's getting more and more 
complicated.