You may be aware when using MUC with memcached and a cache reset occurs, it will cause *all* caches to be reset on that memcached server - not just the one cache store being reset. This is reported as MDL-45375. Understandably, this can be very problematic - the OU detected that this caused big performance blips on their setup (see Tim Hunt's post).
Unfortunately - due to the nature of the way memcached is designed there doesn't really seem to be a good solution to this. One suggestion we had was to use an incrementing cache key prefix and increment the prefix on reset - leaving memcached to clean up old cache keys itself since these keys won't be used. There is a patch on MDL-45375 achieving this, however Eric Merrill has explained on that issue quite extensively why he thinks this is not a wise solution.
So with where we stand at the moment it looks like we will need to close this issue 'wont-fix' and support it with documentation.
Unless anyone out there in the community has better ways to support this use case? Opinions very welcome here or on the issue!