CouchDB offers a different style of persistent storage by storing a flat set of "documents" that each consist of named fields and attachments. CouchDB is highly concurrent and clusterable, and may offer a much faster and more scalable mechanism for storing graphs. It is accessible through a RESTful HTTP/JSON API.
http://incubator.apache.org/couchdb/

Consider that a graph node may be represented as a document, with node properties stored as named fields. Since documents are flat, so the unique document ID would be used for the node's identification property. But to persist the graph, the connector would probably have to store in named fields the parent (no problem) and child relationships (more of a problem for large # of children). See http://wiki.apache.org/couchdb/How_to_store_hierarchical_data for another approach (with issues).