So, you have reverse replication all set up and the usergenerated content is being replicated to all the other publish instances in your CQ estate.
However, upon closer examination of the logs, especially at peak time, you notice that the dispatcher cache is being invalidated too much. And the culprit is /content/usergenerated - the reverse replicated content!
To exclude this, create a user in the system and replicate it to the publish nodes, then modify the rep:policy XML for the /content/usergenerated node to deny this user jcr:all.
Finally, associate this user with the flush agent on the publish instances (the first tab of the user details dialog).