If you look at the kind of things that the search engines are seeking to do with the schema.org vocabularies, you don’t need a graph for any given page. You just need a tree. If you have a page for an event, you put the participants people as children of the event node. If you have a person page, you put the events the person as speaking at as child nodes of the person. Even if in the aggregate things like this obviously form a graph, you can manage with a tree as far as overlaying a piece of the graph over a human-readable page goes. While you can typeset a list of people on an event page as child nodes of the event and you can typeset a list of events a person will be speaking at, you can’t nicely take a larger graph and overlay it on a naturally typeset human-readable page. And this is about overlays – not about larger general RDF models. (The search engines probably wouldn’t store the index they put this data into in an off-the-shelf RDF triple store anyway.)

Thus, the generality of RDFa is a complexity burden. The RDFa advocates who are demonstrating how RDFa can express everything that schema.org vocabularies need expressed are fundamentally missing the point that it’s not about being able to express all that they can demonstrate to be expressible.

A really interesting post about schema.org and the history of microformats, RDFa and microdata. I'm not sure I wholely buy the tree-ness vs graph-ness thing (though I haven't thought about it very hard) but the 'complexity burden' issue is clearly a major one.

Also, there's stuff later on in the piece about the kinds of people who get involved in thiese kinds of activities and why (and how to spot them - or, at least, the consequences of not spotting them).

Great stuff... quite long but well worth reading.