Sending the command "quit" to rec_control can take a very long time to quit- at least a minute, in some cases. This time appears to be proportional to the size of the cache. Conversely, killing the processes with a normal SIGTERM seems to cause it to quit within a few seconds, at most.
This causes a problem with the init.d script included in the package/rpm, which doesn't wait around for the process to stop. Running it with the "restart" option will end up with a stopped recursor on a server with any significant load, as the new instance will be started before the first one quits.
Probably fixed in commit 914.