Bud Gibson at Community Engine has written
a great explanation of the value of allowing users to add their own tags to documents and web pages in the context of a giant intranet. He discusses IBM's intranet system, serving 315,000 users around the world with 3,700 "nodes" (pages, sections?) in multiple languages. He points out that the average person has a working vocabulary of only a few thousand words. Thus, learning to comfortably use 3,700 new ones is totally unrealistic. The best way to facilitate users' recall of pages they have visited in the past is to let them assign classification to those pages that are already meaningful to them. In other words, folksonomy or popular classification, instead of taxonomy or centralized official classification.
That's a big part of the idea behind the new Social Bookmarking tools, but the same concept is also being implemented in many wikis and other systems.
I always say that a big part of searching is conjuring the rhetoric of the writers most likely to have provided the content you seek, but when searching your own archived content there is no reason to repeat that step. Once you've already found something, it only makes sense to make it retrievable afterwards via your own thinking processes.
Technorati Tags (another example!): Folksonomy