Speedup db a bit, but loose some crash resiliency

The loss in durability shouldn't matter, if we can just receive the same
events again after a restart