I think that two of the things you mentioned are working together to increase the speed of the "runaway AI" phenomenon. The new corruption rules and modified FP (where it better accomodates the normal AI's placement) tend to increase the productivitiy and income of not only the AI leader, but also other rival AIs. This means that they can afford to pay full price for new techs, and can do so sooner, thereby fueling the leading AI to research even faster, etc. In other words, he benefits not only from his own improved productivity, but also the improved income of his other rivals.
Re: Armies, I would really like to see some improvement in how (if?) the AI uses armies. I like armies, as an important game concept, that were underdone in the original, and are now more valuable. However, the AI rarely builds them! They seemed to use them somewhat competently when given (i.e. Rise of Rome), but I have rarely seen them in a normal game. Someone suggested making the military academy available without a victory, so the AI could self-build them more easily (assuming they pick the right city to build the academy??). That would help. As for them being overpowered in the hands of a human, I think that is more a factor with Cav armies and a reflection of Cavalry being overpowered for their era. The army then multiplies their inherent advantages (attack and speed). Sword armies give a good advantage, but short-lived. Knight armies are powerful, but a decent match, they can defeat but won't overwhelm muskets. Infantry armies seem quite useful to lead good spearheads for an offensive, or a reaction/reserve force to kill stubborn defenders. Cav armies, on the other hand, are one-city-per-turn blitzkrieg machines. Nathan B. Forrest could only dream of what he could accomplish with a C3C Cav army! Charis' recommendations (limit 2 attacks, limit 3 move) would seem to curb cavarly's excesses, and should be easy to program (then again, I'm not a programmer