I'm really stretching to remember, but I recall that initially the Hurd was planning to have a completely distributed user system (i.e., file handles were essentially URLs, they had a system for dealing with users distributed in the network, etc). They were going to use the Mach kernel message passing in really innovative ways and I seem to recall that they had in mind some fairly radical ideas for distributed processing. I don't think they ever really realized any of that (I haven't actually checked what they ended up doing). My impression was that the project was dogged with a lot of problems, some technical and some not. Had they succeeded, it would have been a really amazing system, I think. One of these days I keep thinking I should check out what they actually did.