Taking up Doug’s human error challenge, I have a couple of incidents that I have been witness to, though note I was only a witness not a protagonist.
There once was a datacenter that required additonal cooling and while this was being installed (in the roof of the datacenter) there was a lot of dust being generated. A bright spark thought it would be a great idea to cover a RAC cluster in dust sheets to protect them from getting too dusty. Unfortunately this person forgot that the servers need to suck in cool air to keep them from meltdown. Admittedly, it was not quite meltdown, but the shiny RAC cluster did overheat causing all nodes to shutdown. A good indication that RAC does not necessarily increase your availability too much. Thankfully, the servers had shut themselves down cleanly and did come back.
Then there was a nice shiny new datacenter that a company was rightly proud of, so decided to conduct guided tours. A dba was showing people round one day, and again decided to point out where a RAC cluster (why is it always the cluster that gets it?) was situated, except this dba got too close to the node he was pointing out. At first the lights of the node were still on but a minute later there was a bit less noise in the datacenter, and one tour was drawn to a swift conclusion.