This almost makes me wonder whether a startup should identify easily-fixed weak points in its architecture and work on the fix so that it's ready to go, but don't deploy it. When the weak point eventually fails, rush to deploy the ready-made fix and then include a wonderful technical post-mortem of what happened, how you overcame an architectural weakness, and post it on High Scalability and submit it to HN. Small downtime; big PR boost.

It's not free if they lost millions of dollars in the process. The % of people in your situation is probably less than 1, so the net is still a loss.