And, if a very bad reality excursion happens, you scrap that branch and restore the last checkpoint. Nobody inside the simulation will ever remember it. You only bother fixing it if halt/scrap/restore becomes a burden.
It probably happens more often than we imagine ;-)
Modern day VM software has various levels of exception checking, and code to catch/mitigate/etc when bugs crop up.
So, a universe-capable simulator might have any kind of behaviour if/when a bug occurs. It doesn't need to be an unbounded, runaway scenario. :)