From pl.atyp.us.

Following storage systems were checked:
* tabled (git clone on 10/27) using boto
* Cassandra 0.4.1 using thrift
* Riak (hg clone on 10/27) using jiak
* Voldemort 0.56
* Tokyo Tyrant 1.1.37 (Cabinet 1.4.36) using pytyrant
* chunkd (git clone on 10/27) using own chunkd.py based on Python’s ctypes module
* Keyspace 1.2 using the built-in Python interface

Results can be found in a spreadsheet, but for lazy ones I want to note, that Tokyo Tyrant was far away from any other concurent (in order of 4-20 times). But since it is single-server storage, it would not be fair to compare against others, which can scale.

Actually I need to say 'could scale', since I did not find any at least remotely similar to fairly scaled numbers, most of the applications behave worse when running on 2-3 nodes cluster.

One can compare them against elliptics network numbers, but getting that it is my results, one can assume it is unfair comparison. I'm pretty sure authors of the all above storage systems had their 'nice' results too.