As discussed on a few occasions earlier, it would be great to have a feature that allows efficient and automatic processing of information over entire subtrees of content. Such a feature could be used to maintain custom reports, search indexes, and other summary information for selected subtrees.
The map-reduce paradigm works great for such a purpose, as you can combine a per-node map operation with a reduce operation that combines the map results of the children of a given parent node. It's easy to generalize this model to cover entire subtrees. And if the reduced information is stored as a part of each parent node, it's even possible to incrementally update the results for an entire subtree when just a part of the subtree is modified.