An implementation of GridFS, such as the one effected by the Java
driver, is really just a set of conventions for chunking files down into
records in collections.
I mention this because wrapping it up in Clojure tightens the binding to
the Java driver, so if you have plans to directly talk BSON at some
point (which would be awesome, and has been mooted for Compojure), you
might not want to add GridFS support yet.
-Steve