When I first learned to put Dublin Core into Web pages, I often found myself replicating data. I would place a DC.creator tag in the head, even though the name of the author was on the Web page. This annoyed me, because I knew that it is bad practice to replicate data like that. When I mentioned this to a workmate at the time, he said that I could probably make a link from the metadata field to the data in XML. At that stage, I didn't understand enough XML to even understand the concept, much less make it work.
Fast forward eight years to DC-ANZ 2005, where Eve Young and Baden Hughes made the point that people updating Web pages often don't update the metadata. One of the problems that they talked about was that metadata in the header is essentially invisible to people editing the page (when, for example, using some wysiwyg editors).
In general, data (including metadata) should be stored in one place only. This prevents drift: if it is only stored in one place, it can only be updated in that place.