You've got something that is inherently an order of magnitude slower than other solutions. Why would you bother to try to scale it? It's not the right tool for the job.
No. CouchDB doesn't support partitioning, so all your replicas need to be able to handle the full write volume.
It will soon. We have an 8-node cluster running in PROD right now. A Dynamo-like layer provides consistent hashing, partitioning, read/write quorum constants and the like.