Comment by oneshadab

You'd lose durability against process crashes.

If your system has a reasonable tolerance for power failure (multi-az multi-cloud), this can provide much better throughput