If we imagine this is sim after the style of a game like Civilization, where the player is measured along multiple metrics of success and has competition from multiple other factions, I would substitute P(rel|sim) with P(rel|gamesim), the probability that religious veneration is one of the metrics in the game, but I would assign P(gamesim) a much lower probability than P(sim), because it occupies a very small part of the probability mass that I assign to a simulation. As before, a large part of the improbability would come from locating that particular type of sim in possibility space.

I would also have to revise P(follow-thru) way down, because more than in the other scenario, it seems like it would be a complete waste of time, completely tangential to the purpose of the simulation and not at all worth the computation time it would take.