More AI complexity would likely increase the time to play the game, it won't necessarilly increase minimum requirements. It would just take longer for the AI to process its turn (so they might recommend a faster CPU and such).
The goal of the Task Force Strategy is to eliminate what I think is a weakness in their program (every unit has a decision cycle) by simplifying the amount of decisions the AI has to make.
Trying to intelligently guide and do the pathfinding for 100 units would take 200 'cycles' (one cycle to do pathfinding, one cycle for decision making). Splitting those 100 units into 4 task forces of 25 each would take 104 cycles (one decision cycle per task force).
In reality its not so easy (pathfinding time is different than the time involved for a decision), but this shows quickly the reduction in work needing to be processed.
As it is now, I believe the AI processes a target for each unit in its army. It decides which is the weakest unit that it can attack and kill. This leads it to send 20 infantry on a quest to defeat a lowly cavalry. Indeed, I've seen it send 30 units away from the front for a horsemen once, and then later in the turn the horsemen was killed by a unit that was close to it to begin with. So those 30 units didn't even have to move back to kill that weakest unit.
If that unit didn't exist close to my horsemen I believe the 30 units would have continued traveling to my little horsey, used however many it needed to kill it, and then turned around and sent the remainder back to the city target.
The task force would change that. The primary division would be a stack (for simplest version of this plan) and it will be set to a NON-UNIT target, such as a city, resource square, or fortress. Its goal will be to claim that target with whatever force necessary, taking the path of least resistance (resistance defined as a mix of time and simply killing units that are in the way). Since each unit is no longer making a decision, merely pathfinding, it doesn't have to spend the time looking at every enemy unit on the map. This time saving can counteract whatever increased decision time is required to have the task force make an intelligent decision.
So the tradeoff is thus:
Each unit takes less time to process a turn.
Each stack (task force) can take five/ten or even twenty times longer to make a decision, so it can make an INTELLIGENT one.
This is compared to the AIs relatively dumb SIMPLE UNIT TARGET decision which takes less time.
The net result is less overall decision time, because although the simple unit target is both easier to program and takes less time per decision, it is not efficient for large scale armies as the combined decision making. The other net result is that the decisions made will also be more comparable to those made by a human player (i.e. humans tend to be goal oriented, not unit oriented like the AI).
I would predict that the actual work done by the hardware would only increase slightly, if at all. You'll be surprised at how much computing time can be saved from approaching the same problem with a different strategy (I'm a programmer btw).
The pathfinding problem still exists, but the advantage of the stack is that if you use a strict stack control (i.e. every unit starts at the same place and ends at the same place and moves at the same speed), you can replicate the pathfinding pattern for one unit and simply copy it for all the other units. Therefore you might even see a decrease in computations for pathfinding purposes (freeing up more time which can counterbalance the time being spent on 'strategy'). REgardless, as it is the computer processes a pathfinding cycle for every unit every turn. It can't possibly get any worst through this innovation, and it has the potential to be better.
I won't go into other facets such as how the concept can be applied towards AI production so they don't make stupid unit decisions or into the more complicated ideas of programming an AI so it not only knows how to attack (with a massed task force), but where and when to attack (i.e. the devious ROP scheme humans love).
MSGT John Drew is correct in that stacks doesn't solve everything, there are still weaknesses involved. But I think further improving the AI to properly utilize/defend artillery, realize advantages of speed, defensive territory, and making formation stacks (i.e. not all on one tile but arrayed in a loose collection of stacks) can be done easily with this idea.
Also coordination between stacks will become easier, make navies and air forces into task forces as well and make combination strategies between multiple stacks. If the AI had to make a decision for every bomber it might just randomly bombard your territory (as it does to me). If the AI air force had a mission to say, cut off reinforcements to a target city for one of the ground forces, it can quickly recognize a line of roads it should bombard and quickly make a fast decision cycle for every bomber (bomb each square until its road is gone and move to next square). Coordinating stacks takes so little time compared to trying to gauge a situation involving lots of different types of units.
I already play with a task force strategy myself, coordinating my ground forces into 'armies', my ships into 'navies', and then assigning airplanes when I get them to corresponding regions or navies (with carriers). Its proven incredibly effective in winning battles, so effective that I feel a little sad. The AI usually puts up a good entrance fight at one point, but once I develop multi-pronged attack strategies with multiple forces and bombardment support? Well it gets just too easy. Even with relatively small forces (say 20-30 units ground, 20 air/navy support), I'm conquering dozens of cities in a few turns while isolating their main army on a single front.
It's gotten so that I don't even call in AI allies and declare wars on multiple people at a time so that I can have at least some semblance of a 'multi-front' war where the single target mentality at least is hidden by the fact that five other countries each might have a different single target.
But if the AI could even do only a weak imitation of my general strategies it would be infinitely more dangerous to fight. If I knew I had to contend with task forces instead of a trickle of units my appreciation of the game would increase with the need for improving on my strategies.
(comeon, Catt, hit my 
)
) game under 1.29f, I felt compelled to play an almost completely peaceful game - after paying an arm and a leg for a source of foreign oil, I was very happy to see a new source spring up in my territory - sorry Montezuma, I don't necessarily want to renew that tech + 75 gpt / oil trade deal we had 
