A Document Runner would provide functionality to run a transform function across all documents in the database. This way you could do maintenance and migrations, with incurring a bunch of http overhead.
We may not be able to guarantee that runner functions see each document only once, so functions need to be aware only to modify documents that require modification.
Changing the format of a timestamp across all docs
Removing sensitive fields before replicating to another jurisdiction
(View runner... slightly different) Materializing the results of group=true reduce queries into a dataset for further map/reduce processing (a common pattern in Hadoop)