I admit to liking Riak quite a bit. The test results seem to show it being quite a bit slower (which is a bit disappointing), at least in the 4 node configuration they used — riak recommends 3 or 5 nodes as I recall. They also used EBS volumes..which is.. unfortunate. We have had extreme variance with our EBS volumes in the past. We had to move to guaranteed IOPS EBS volumes (raid0 strip across 8 of them) to get decent performance out of mysql when we were using it. That could add a huge amount of variance to their tests. It seems like they were using ebs-optimized instances, but I don’t see where they said they were using dedicated iops EBS vols.
It would have been interesting to see how the various NoSQL dbs under test behaved in a degraded state (pull the plug on one of the servers in the cluster). Would also have liked to see couchbase in there.