General Observations of 2.0 on aws
beam.smp is acting very volatile. With a default bucket and no data and no traffic it pings the cpu anywhere from 3-10%. Running a load script (creating keys,reading them back) the cpu time spikes and causes AWS to suspend the instance, granted this suspend behavior is effected by your instance size, but its just not promising in terms of scaling. The suspending seems to make things worse, it causes the beam process to stay high for a very long time. Then it being high causes more suspend. You can see the cycle continue for a long time, it takes a while for everything to get stable. This is with no new traffic added to the situtation.
As a comparison couch 1.8.1 even under the highest load running along side a webserver on the same instance the max cpu for beam and memcache seems to be 4% beam will take 2% and memcached will take 2%, the rest of the cpu under load is going to python processes serving the web requests.
According to the new feature page it looks like a lot of things have changed. It may be that 2.0 is better on real hardware, but I wanted to chime in for AWS users. I'm also relatively new in using couch, but for the time being it looks like we're going to launch with 1.8. The big feature for us on 2.0 would be the views, KV stores in general aren't fast at keeping concurrently active lists.
Initial test with a single node, I did another test with 2 nodes but I observed the same behavior.