Structures are useful - I'm actually a lot more comfortable putting things in MongoDB when I know they're coming from a strongly typed language than if they're coming from Javascript, because that way I know they've gone through a rigid "schema" at some point. But I don't think SQL's very limited datatypes are the best way to express the constraints on a typical business object. What I'd really like is a database that stored object instances, in a language that made a type-level distinction between serializable/value objects and non-serializable objects (one of those projects I keep meaning to do is the design of such a language). ZODB tried to do this in python, but it'd be far more useful in something strongly typed - I've heard Google have had some success with storing protobuf objects directly in a database? The protobuf schema is rigid enough to provide the kind of guarantees you get with SQL, but a lot easier to map domain objects onto.

Which really sums up the advantage of MongoDB to me - JSON is a more natural fit for domain objects than SQL tables.