I’ve seen cloud slowness create weird Stockholm syndrome effects, especially around disk latency.

It always makes sense to compare to back of the envelope bare metal numbers before rearchitecting your stack to work around some dumb cloud performance issue.