It's possible, and easy to some degree, to implement a triple store on top of an RDBMS. There are several systems currently available that do this with varying degrees of success. However, they tend not perform all that well due to transitive self joins their design usually requires. This is why serious vendors who are providing a triple store backed by a relational database, such as Oracle, have customized handling to help improve their efficiency in these situations.
In my experience, native triple stores, those designed for the purpose of storing and querying RDF, always outperform solutions shoehorned on top of a relational system. So while they're very much databases and have a lot in common with a traditional RDBMS, there are still design choices in their implementation that makes them better suited for answering SPARQL queries.