actually, we have a cluster with about 20 nodes or so, but it runs gridengine, with shared filesystems. The opposite of MapReduce ;)