Cloudant has a very cool couchdb fork that uses a dynamo style k/v layer that then uses couchdb as its storage engine. They implement chained map/reduce with the caveats that A) the chained m/r is generated async, so it is always out of data a bit and B) I never got it working in their hosted product. I checked them out for the chained M/R but was really wowed by the scale out capable with their system. The dynamo layer takes care of sharding, migrating shards, re-reducing parallel queries etc.