Partitionning a RDF DB by molecules

I would like to have good performances when dealing with Roads+Towns in dbPedia.

To achieve that, I plan to:

1. extract all Roads molecules from dbPedia. (a Road molecule being "any resource of type dbpedia:Road + its label + its longitude + its latitude + its thumbnail").

2. extract all Towns molecules from dbPedia, in a similar way. (a Town molecule being "any resource of type dbpedia:City+ its abstract + its longitude + its latitude + its thumbnail").

3. load all those molecules in a dedicated DB (RDF store, or NoSQL database [1]).

4. query that dedicated DB for Roads+Towns information with acceptable performances

I would like your opinion on such a partitioning strategy.

Will it really improve performances to query a dedicated DB, vs running the same query on the full dbPedia DB?

Do existing RDF DBs already provide a similar feature (afaiu, this feature is called "rdf materialized views")?

[1]: my molecules have no cycles and can be serialized as trees. So NoSQL is an option for storage, indexation and retrieval.