You should only create a new vocabulary if there is no (popular) existing one that could be used.
Dublin Core, FOAF and SIOC are popular ones that almost every forum could use. However, I'm not sure if these can be used with microdata (I guess it should be possible, but I don't know microdata very well). But they work with RDFa, which is very similar to microdata and a W3C recommendation (like HTML). RDFa can be used in HTML5, too. If new to RDF, you'd probably want to use RDFa Lite first (it has prefixes for DC, FOAF and schema.org vocabularies pre-defined).
I wondered about the impact this may have server performance, being hit by all those spiders crawling the vocabulary.
I don't think there are many (if any at all) microdata crawlers that would try to visit the vocabulary URIs. In most cases they wouldn't find any content there they could make use of, because the vocabulary URIs work as identifier only most of the time. Even if such crawlers arise, you'd hardly notice any performance impact because they'd probably cache it anyway.