I think this type of rubber-banding sounds really interesting, especially if they can get the AI difficulty is determined primarily by how the AI plays rather than by cheats and bonuses. That being said, When you hit that tipping-point where you know you are going to win, it may be tough to raise the difficulty enough without it feeling really unfair to the player.
I don't see the suggestion as proposed as a rubber-banding type thing - it's the player who is deciding the difficulty should change, not the computer. The player's goal in changing the difficulty during an in-progress game is most likely to keep the game fun, whereas the goal of a dynamic difficulty system is to keep the game challenging; these two goals are not the same thing, despite what some people will tell you. Worse to me is that I personally find that dynamic difficulty can make progression less rewarding, simply because it means that I haven't really accomplished anything despite all the stuff I've done, since my character or empire still has more or less the same strength relative to the same opponent in the 'after' picture as in the 'before' picture (when done particularly poorly, as in TES IV: Oblivion, you're actually going backwards in relative strength compared to the same opponents as you advance the game).
Dynamic difficulty done well can be a good thing, such as by introducing thematically-appropriate tougher opponents, as when done well it can provide a challenge without that sense that you're not making any headway despite all your efforts. When it's not done well, it can make you feel cheated or like you've made no progress or even gone backwards (why is it that in Oblivion my level 1 Bosmer Assassin can kill a goblin in 1 shot with a bow but my level 10 Bosmer Assassin cannot under the same circumstances with level-appropriate equipment and given the same target?), or like the computer's an idiot (if it has the economic power to sustain a fleet bigger than my own despite having only a hundredth of the colonies that I have, why did it never make use of that economic strength while it was in a position of strength. Ultimately, though, dynamic difficulty is not a solution to the tipping point issue; at its best, it delays the onset without irritating the player, while at its worst it makes the player feel cheated.