Just promise yourself that you will never try to map a relational data model to a NoSQL database like MongoDB or CouchDB... This is the most common mistake developers make when evaluating emerging tech.
That approach is analogous to taking a car and trying to use it to pull your cart down the road like a horse.
It's a natural reaction due to everyone's experience of course, but the real value in using a document database is being able to simplify your datamodel and minimize your suffering as a developer. Your codebase will shrink, your bugs will be fewer and easier to find, performance is going to be awesome, and scale will be much simpler.
As a Joomla founder I'm biased :-) but coming from the CMS space, something like MongoDB is a silver bullet as content maps very naturally to document systems.
Another great case for MongoDB is real-time analytics, as MongoDB has very strong performance and scale particularly regarding concurrency. There are case studies at the MongoDB.org website that demonstrate those attributes.
I agree with the notion that each database has its own aims and use cases; take the purpose of each database for evaluation accordingly.
