One weakness of tags is that the same tag can mean two completely different things to different people. For example, a recent memoir by CNN correspondent Anderson Cooper is tagged by various Amazon users with "news memoir," "blue eyes," and "hunk." Since "hunk" is a tag with many possible meanings, it appears on many products with seemingly no connection--like a movie starring Russell Crowe, the DVD Forrest Gump, heavy sweaters, and books about the "chunky" clothing style.
Conversely, various people will assign different tags to the same thing--one person may tag photos of their dog "cocker spaniel" while another user tags the same photo "canines." A search for the tag "dogs" might not turn up either photo. With books, an Amazon user may assign the tag "Christmas" to a book about baseball, meaning that she intends to buy it as a Christmas gift. Meanwhile, customers using the tag "Christmas" to search for Christmas books will be frustrated.
Like any valuable tool, tags can be abused, too. If tagging goes mainstream, spammers will try exploiting tags by adding their irrelevant tags to popular items.
Tags aren't necessarily linked with semantics. So the word "blow" could be used as a tag for wind, cocaine, sucking, breath, or a picture of a tornado, or the sound of air rushing. The user of the tag, not a search engine, decides how the meaning fits for them.
Advocates of tagging assert these fears are overblown. With enough users, tags become self-correcting, so inappropriate or useless tags will be drowned out by the good ones.
Whether it's Amazon, LibraryThing, or perhaps Google, whoever builds the biggest collection of tags will have an amazing insight into how people think about information, and will have a important tool for bookselling--the most detailed, current, and useful book index in history.