I believe their point is that deploying, for example, four types of NoSQL database creates four, separate sets of hardware etc.
i.e. it is not that NoSQL systems are inefficient individually, it's that it is not efficient to maintain four separate resource pools.
(This is still possibly FUD nonsense: Oracle has to exist as an independent resource pool only because their virtualization licensing requires all potential CPUs to be licensed. Are any/all of the commercial NoSQL implementations licensed this way?)
It takes less time to configure an HBase cluster than a single Oracle server under Linux, and even then you can just create a single HBase image to deploy in hundreds of servers.
i.e. it is not that NoSQL systems are inefficient individually, it's that it is not efficient to maintain four separate resource pools.
(This is still possibly FUD nonsense: Oracle has to exist as an independent resource pool only because their virtualization licensing requires all potential CPUs to be licensed. Are any/all of the commercial NoSQL implementations licensed this way?)