This. Failing a public defense is not like a startup failing, it's more like an IPO failing (as in, stock goes to zero within a short time). You don't do it if you aren't 99.9999% sure it won't fail.
I've seen it happen five or six times (usually it's the same group of professors) at a top-tier research institute.
I've seen a number of things happen during a defence, but not this. The worst was seeing two professors (one of whom was the candidate's supervisor) on the committee argue with each other while the candidate just had to watch. It took a lot of work for the chair to reign them in. I think it was difficult for the chair because they were arguing in German and the chair didn't speak it since he wasn't from the German department (my friend was getting her Ph.D. in German literature).
The reason why failing a PhD thesis defense is rare is because your adviser should prevent you from defending until the product is sufficiently strong. Many, many, many more people fail by dropping out of a PhD program during their dissertation than fail by doing a poor defense. Comparing successful thesis defense rates to startup success rates is not valid.