That sort of thing is actually causing spikes in load on the cluster. I think the quickness of the requests must skip the cache. So it's a bug really, I just need to detect recursive URLs and extract the deepest link.