Comment by badatnames

> this can be a matter of scaling at 50% capacity

At which point we're paying double just to work around cold start latency, which was the original point.