I've been thinking about this. I estimate a few people per 1000 would do an atrocity for no reason if they were guaranteed no consequences, and the deaths if the switch is pulled are 2^(n-1)^ for the nth switch. The expected deaths will cross 1 somewhere in the high single-digits, then (since it's outcome*chance), so the death minimising strategy is actually to pull yours if the chain is at least that long.

I find that counterintuitive, because the overwhelming most likely outcome is still no deaths if you let it go. Humans, myself included, just aren't good at tracking large numbers of things like victims intuitively, I guess. a 1/1024 chance of 1024 deaths feels like less of a big deal that 1 guaranteed death even if I would maintain that it's not.