Hi,
I've been testing out couchdb, and came across the concept of views.
Views in couchdb are keeping a pre-calculated map-reduce on all data in one database(collection), and they are updated automatically on every change on the original data set.
Knowing that mongodb supports mapreduce, i think it would be a good idea to be able to set up mapreduce that would output to another collection while keeping the mapreduced data synchronized based on inserts, updates, and deletions of the documents in the original collection.

Often there are times where mapreduce is necessary to be used, and keeping the data set synchronized without the hassle or delay of the current incremental mapreduce method will make lives easier.

For details about the implementation in couchdb, read the following articles:
http://horicky.blogspot.com/2008/10/couchdb-implementation.html (sections Views Indexes and Incremental View Update)
http://damienkatz.net/2008/02/incremental_map.html