Nice article, but the replicated approach isn't exactly comparing like with like. To achieve the same semantics you'd need to block for a response from the remote backup servers which would end up with the same latency as the other cloud providers...