I believe the objective is to replace expensive RAM with less expensive SSD if your workload is suitable for it. It appears there are workloads for which that is an acceptable cost/latency tradeoff.