@Ayende asks what the sweet spot for RavenDB would be.. The thing is a lot will become more clear with experience using it. Right now, I envision a small to medium sized app that doesn’t have an overly deep object model and doesn’t require a lot of inter-joining queries (because that would mean a lot of map-reduce queries).
But RavenDB is scalable and seems to have good performance already so it should be capable of handling large data systems. I just hesitate with any new database until it actually proves it can handle really large data stores. You see, really large data stores have their own set of problems with performance and scaling which you don’t see in smaller systems. Try dropping, reloading, and full-text indexing a multi-terabyte sql server db every 24-48 hours while keeping the data driven site up. You’ll see what I mean pretty quick.
I have my reservations, but am otherwise very eager to use RavenDB in real applications. It has a lot of promise; I even started a demo app called RavenBlog. I find it really let you focus on business rules and implementation because you don’t spend nearly the time determining the persistence model. And that is a VERY good thing. (which is true of NoSQL in general)
So to summarize my answer: small-medium size apps with limited object models, but I think it could prove to handle bigger systems as experience proves out.