It came to me while reading
this paper, that during world-gen (both initially and as part of the ongoing world during post-play) there were only changes to the agents within the world, not to the world itself. The landscape is pretty much baked in as-is so long as it isn't deformed by agent-based actions (races build sites, monsters make lairs, ect...).
Usually, the complexity of the sim crashes the game after 2-300 years at most or anything more than a few dozen cumulative sites. For the amount of world generated, this limits the types of world histories that can be simulated, making a potentially deep history artificially shallow. I suppose that a "brute force" approach could be used by just throwing more efficient chipsets and RAM at the problem, but that would merely push back the problem to another world-crashing threshold. Instead, it might be better to work within the constraints provided by the system and create a balancing effect that will temporarily remove complexity from the simulation: A "Natural Disaster" like a new volcano, an earthquake, a flood or drought - a global change of climate that creates widespread devastation... all of these would be able to make widespread changes to whole regions, killing off agent groups and removing whole swaths of populations - which in turn would free up map space for those parts of the world and also allow for whole sites to be relocated to caverns as ruins for instance, or to be covered in magma and buried like Pompeii. Periodic mass death or site devastation would make the world not only seem more dynamic, but also would free up resources that could be used to simulate the non-affected agents in the simulation. If the trigger for the natural disasters was based on a threshold of system resources, was based on the machine running the program, then it would scale according to need. Histories could be long or short, civ numbers could be few or many, and the population of caverns (and crevices/canyons allowing for those caverns to interact with the world above), would all provide a richer experience without needing to add much in the way of newness to the already generated landscape. Also, this would make world crashes less likely, which is an awesome thing in my book.