Looking at the benchmarks, something looks wrong in the graphs. For example, for one it looks like they were doing from 0-2M ops for the test. Some of the engines were claimed to do 1M ops per second. That would result in a 2 second benchmark?
Is a 2 second benchmark actually useful?
OpenLDAP's MDB is also in this space that is new enough that most people haven't heard of it's exitence.