but it would probably still lead to some frustration at the threat-ball becoming a threat-wall.
That's why the overall adjustment to how much firepower ratio the AI waits for before attacking would be needed if how much the AI penalizes firepower of turrets is reduced.
In theory I could have it only count stationary non-snipers that are actually within range (say, range*1.2 to account for the fact that the AI needs some room to maneuver) of the relevant wormhole for the purposes of "do I go through?". That would be some additional cpu cost but it would be on the AI thread and it's ok if it slows down a little (the "mistakes" that the AI makes are not typically caused by slow cycles, etc).
In that case, it could consider those turrets as having the full firepower
When I first posted my mantis issue, this sort of dynamic turret firepower adjustment came to mind. However, I now realize how finicky, and in the right hands, exploitable that would be. I think a properly chosen denominator for a better "rough approximation" of "turrets being stationary consideration" may work just as well in most cases.
Then again, this sort of dynamic lookup would allow configurations like Nodor described to still not hold off threat from coming through very much (though, maybe with a slight adjustment such they come through in slightly larger "batches"), allowing people to easily "bait" the AI with unusual turret configurations at the penalty of not having as good "alpha strike" damage or "critical point defensibility", but still keep the AI from sending in 5 ship "waves" into a very aggressive turret ball right at the wormhole they are entering.
This way, if people are willing to risk placing turrets such that the AI could possibly slip past them, the AI would be more willing to come in and try to do just that, allowing people to still easily "bait" small threat balls. But for those who place turrets in such a way that the AI cannot slip past them, the AI will consider that too and wait for a bigger buildup to stand a chance.