That's more like a key-value store. One big problem: For any operation on an 'object', you must download the entire object. (I.e. let's say each object is a blog post with metadata and all comments. You would have to download all comments for all blog posts just to list the titles of the blog posts. Ick.)
With a docstore, you can have the database scan all the objects and pick out the field you want. It saves a ton of bandwidth, and client-side processing. (Mongo saves a lot of server-side processing by using BSON for disk storage and wire-protocol.)