Firstly, an as has been pointed out before, owl can be serialised in RDF.
Secondly, OWL adds ontological capability to RDF (which on its own only provides extremely limited capability for formal knownledge representation), by providing the apparatus to define the components of your triple using formal computable first order description logic. That is what posters here mean by when they talk about "semantic richness".
Thirdly, it's important to realise that in OWL-Full (for OWL 1) rdfs:class and owl:class are equivalent and in OWL-DL, owl:class is a subclass of rdfs:class. In effect, this means that you can use an OWL ontology as a schema for RDF (which does not formally require schemata).
I hope that helps to clarify further.
