The idea case is of course to mark posts with one or more labels. And let the users subscribe to one or more labels.
This is specially true of sites such as blogger or Always on which have many members and publish multiple blogs.
The algorithm for selecting blogs will now become bit more compicated; rather than just pulling in the URL of feed, you would use a subscription-wizard to browse through a list of blogs and subscribe to the ones matching the labels you are interested in.
But is there more practical use for this? With so much of information around, labeling and selection based on labeling are going to be key issues. After all, labeling is not necessarily efficient when performed at the time of creating / accessing content; it is done when content is found useful - typically when you search for information in particular context. And such a labeling process is performed assuming the information will be useful later, and it may not.
But collaborative labeling is indeed useful: If I label some information, it would be useful to everyone else. But what should be the infrastructure for it? For example, gmail allows labels to be put on emails. But: (1) I need to create my own list of labels; can't borrow it from some generic list available off the web, and customized by me (2) I can't send this labeling information along with email I create, so it is duplicate work by recipient (3) If the sender adds new labels to a thread of communication after emails were received, there is no way to share them.
Of course, automatic categorization and better search techniques might remove the need for labeling in the first place.
Perhaps more debate is needed on these lines.