There is a recently open-sourced fork of CouchDB that does the sharding thing. With this addition to the capabilities, CouchDB really can scale up to hundred machine clusters.
I should make clear that the notice is titled "Notice for 1.0.1" because the 1.0.1 release (due this week) will contain the fix. The bug is confined to 1.0.0.
the benefit of a bulk index is the ability to shove everything into a file and fsync once. so splitting across files won't really give the performance benefits.
I've been in production without issue for years in some installations. I probably could have raised the "production-ready" flag in 2008/2009 if I held myself to the same standards as a lot of projects.
I've got truly offline replication (as in you can run me on Android and your laptop and keep them synced via the cloud).
Also, none of the other NoSQLs are really set up for schemaless queries like I am. Incremental map-reduce lets you normalize a bunch of different heterogenous document structures in your view, which is more flexible than the key-path indexes Mongo has, and more real-time than Hadoop/Google style map reduce.
http://blog.cloudant.com/cloudant-core-is-open-source