The Semantic Web seems finally close to maintaining its promise about a real world-wide graph of interconnected resources. The SPARQL query language and protocols and the Linked Open Data initiative have laid the way for endless data endpoints sparse around the globe. However, for the Semantic Web to really happen, it does not suffice to get billions of triples out there: these must be shareable, interlinked and conform to widely accepted vocabularies. While more and more data are converted from already available large knowledge repositories of companies and organizations, the question whether these should be carefully converted to semantically consistent ontology vocabularies or find other shallow representations for their content naturally arises. The danger is to come up with massive amounts of useless data, a boomerang which could result to be contradictory for the success of the web of data. In this paper, I provide some insights on common problems which may arise when porting huge amount of existing data or conceptual schemes (very common in the agriculture domain) to resource description framwork (RDF), and will address different modeling choices, by discussing in particular the relationship between the two main modeling vocabularies offered by W3C: OWL and SKOS.