I'm currently working on generalizing our backend implementations - it will help with benchmarks of these new DBs.
With this work I hope to lower the barrier for new backend implementations, so they can share most optimizations if they fall into one of existing data models (KV, NoSQL, SQL).
KV is mostly done, so expect to see new LevelDB port, improved Bolt2 performance, and maybe an experimental CockroachDB KV backend (raw KV access, since CDB is basically a distributed KV).
NoSQL and SQL will follow shortly. Once NoSQL is ready we may try to benchmark these different solutions.