Maybe its all the circuits

quote: AI bots that are close, the same level, and on the same team do seem to merge up, so I wonder if that's used to free up processing power

Most of it, but not all.

quote: One has to wonder how players that host those large team matches can cope with the load that comes with so many players participating in a single match

Games don't have a single host (not just the large team games, all of them -- including yours and this one). All network data is relayed between all clients through the ZK server and everybody runs the simulation on their own. All hosting a room does is give you some control over the setup.

Bots do have a host who pays the entire cost of running it, but this doesn't have to be the room's host. You can split the burden with others as long as you make sure they have a good CPU (having someone with weak CPU host a bot would be pretty disastrous).

Players by themselves cause almost no load, on the other hand the number of teams has some effect. Most of the load comes from units and bots. Large team games have just 2 teams and no bots; and quite often they are played on small cramped maps to limit the number of units as well.

As far as I can tell the performance drop you're seeing when adding more players is because you add some bots to match the number of teammates in each team, which has some minor cost despite the mergers and more importantly because the added starting resources make the game escalate to larger unit counts quicker (which also increases the cost of the bot handling those units).

I see. Thanks for the insight. It might be more worthwhile limiting the number of bot teams after all.