What is this -10 modifier of which you speak? I have scarcely any recent experience of the AI achieving alliance so how to break one is somewhat shrouded in mystery.
Lets say the AI accepts a mission against its ally because of some vast bribe. Firstly I don't understand why such a mission would be carried out, as the diplomatic penalty for failing the mission would be far less than for breaking a treaty. So I suppose that there would have to be an immediate relations penalty for accepting a mission against an ally, and of course there is a penalty for breaking the treaty.
If the diplomatic actions penalty modifier was then strong enough and did not degrade over time, then the former allies would remain apart. If it degraded over time it would depend on the size of the penalty and rate of decline. From what I understand, the subsequent combat actions do not represent enough of a penalty to keep the former allies apart. The negative cap might be quite high, but it seems doubtful that the penalty would ever get anywhere near the cap.
Of course if the diplomatic penalty for breaking a treaty was sufficiently high and did not degrade, then former allies would never make up.
That's what I know of the theory of it, anyway. I'd prefer a system where an ally realised that its partner was getting too strong or had too many planets near its borders and over time it became less keen.
If you set up a singleplayer FFA game where several powers can achieve a joint victory you are playing a very odd version of the game, no?