Hi,
spec by example is a conversation technique, so if this is a case that is important to discuss, it should be part of the conversation at least. the fact that something is ultra-rare makes it an interesting boundary for a conversation, then you can later decide if the risk of that actually happening makes it worth to automate a test based on it, or perhaps design the system in a way that de-risks that in some other way, so that example becomes covered by some other key example.
if system recovery is a critical feature you want to specify (and potentially test in an automated way), perhaps this is a signal telling you that you should isolate that module so it can be kicked off easier and validated. automated tests don't lead directly to good design, but they can lead you away from bad design (eg if it hurts to test something, it's telling you that you should modularise the system a bit more).
gojko