Otherwise it looks like your server is running out of threads to handle requests. Clustering would help very quickly - isn't the clustering package in use in a few of the big .LRN sites?
(And fixing those long-running queries :)
