While building TouchDB Jens documented the replication protocol here: https://github.com/couchbaselabs/TouchDB-iOS/wiki/Replicatio...
I am working on (actually am right now) the in browser implementation of CouchDB called PouchDB (https://github.com/mikeal/pouchdb) it is nearing a beta release.
http://www.dataprotocols.org/en/latest/couchdb_replication.h... https://github.com/couchbaselabs/TouchDB-iOS/wiki/Replicatio...
From the end of http://wiki.apache.org/couchdb/Replication
I'll see what I can do about getting a good publication of the replication protocol. People have been asking about it for a long time. I've only come to fully grok it late last year when I tried to write PouchDB and I talked through it with Dale when he picked up my slack. We can probably get that together.
A great first example that I'd love to see is using PouchDB to implement Backbone.Sync for a Backbone app.
Its very basic but has a working Todo example.
The focus client-side is on speed, battery-life, ease of programming, and perhaps most of all, maintaining separation of concerns between the server-side data model and the client data model. The server must be free to evolve independently of the client, and many remote clients will run legacy versions of your software for extended periods of time.
Moreover, rarely is data distribution sufficient to model high-level interactions between client and server, which is more likely to introduce bugs, complexity, and significantly detract from the long-term maintainability of the client/server protocol.
The end result is that you don't particularly benefit from CouchDB client-side, and especially not replication -- you need a well-defined supportable network model.
As far as the maintainability concerns, it's an interesting point. There are certainly apps which may want to sacrifice the ease of use and performance advantages of having sync done for you, because they have specialized data sharing requirements.
But for a whole lot of apps the benefits of sync far outweigh the contraints of having the same data model on the client and server.
Some example use cases where sync can simplify programming (or even make a class of apps possible which would be much harder without it): point-of-sale, retail, medical records (doctor with an iPad and limited wifi), military (need that data no matter what).
Here's a pdf of the slides I presented yesterday at Chicago's WindyCityGo conference: http://dl.dropbox.com/u/14074521/syncpoint-windycity-small.p...
You're just moving the costs around, and likely transmitting far more data than needs be sent.
In addition, you're front-loading the transmission of data that users don't need right now.
Having built many network-facing mobile applications with exceedingly large user bases, I'm unconvinced. The maintenance complexity, the excess transmission of data, issues of data consistency over time, the lack of integration with client-side business rules without considerable investment in code that significantly replaces the sync logic.
I think that the reality of the situation is that in the rare cases you need offline, fully synchronized data, it's not particularly difficult to just download what you need as an atomic unit and save it. The majority of the time, however, on-demand request handling is more efficient, well-defined, and maintainable.
As far as ease of writing apps (and maintaining them), time will tell, some of our users have been at it for a few years now. The things I've seen people do, show me the synchronizing document model works.
[edit] To be clear - Couchbase Server is mostly used for the explicit network model you advocate. For instance, the OMG POP mobile success story is using Couchbase as regular backend database storage.
To me, that'd be a much more compelling reason to use CouchDB, since syncing between server and client is a pain, and having CouchDB on both servers and client would make development life much easier for me.
If not, then publish the replication protocol. I tried looking for it once, and only found scant information on the mailing lists and documentation.