I have just started reading Organising Knowledge: Taxonomies, Knowledge and Organisational Effectiveness by Patrick Lambe. Of course, I turned straight to the last chapter – where folksonomies are discussed. Lambe argues that folksonomies work best with large quantities of new content, where social tagging creates some way of grouping similar items quickly and cheaply, but when the users start to demand comprehensiveness and accuracy in their searches, and once the size of the collection becomes too large, some sort of formal taxonomic structure works best. Sites are starting to add traditional facets, like location, to control and focus their social tagging.
There is a counter argument that for very small well-defined communities, social tagging works well, because the users have a good understanding of the terminology, tend to think in the same way, and so tend to use very similar tags. This would explain why the folksonomic approach was so popular in the web community – a new highly specialised community all speaking the same jargon were all tagging new content in very similar ways. The danger is that once the community expands, people stop using terms with such precision and the helpfulness of the social tagging get diluted.