I wouldn’t say subset per se, more tangential.
Looking at experiences with the SPARQL plugin for Neo4J it’s much slower than a dedicated RDF Triple Store.
The other niceities you get with the RDF model is identifying everything with a URI, making the merging of datasets trivial (in the best case), and formal ontologies which provide various predefined models for building your data, e.g. FOAF, Dublin Core Terms, SIOC etc.