Hm... interesting, I had the opposite experience. It was the "aha" moment for me when instances on us-west-1a and us-west-1b were rebooted at different times, which let us avoid any downtime.