Given what my experiments (which I'm going to share herein) have turned up, I've decided that even this sub-topic of "flak is/isn't overpowered" deserves its own thread.

I've tested equal fleet supply of Cobalts vs. Gardas in the small scale, i.e. 4 Cobalts vs 5 Gardas. (Keep in mind that these two fleets have quite unequal production costs though, the Gardas being considerably more expensive; for an equal production cost, the numbers would have to be reversed.)

My experiments and results are as follows:
1) Well microed Gardas vs. so-so microed Cobalts: Gardas won with 2 losses. What I mean by well microed Gardas here is: turn at the right distance, so that the Cobalts are in range of the Gardas almost all the time. Run straight at the Cobalts' formation to maximize time spent in their blind side; this actually requires a fair bit of continuous attention and adjustment, as you'll see from the next test. What I mean by so-so microed Cobalts here is basically not let them focus fire to the point that they raise any enemy's mitigation above 30%. The Cobalts also stayed in one formation, but attempted to foil the Gardas' attempt to continuously blindside them by turning half their formation clockwise and the other half counterclockwise. This is actually quite important (under the circumstances of this test where Cobalts were bunched together), because when there was only one Cobalt left, the Gardas managed to blindside it for nearly a full minute (by just moving in around it at the right distance), and during this time the lonely Cobalt basically didn't manage to fire at all. 
2) So-so microed Gardas vs. barely microed Cobalts. What I mean by so-so microed Gardas here is basically an attempt to mostly automate their maneuvers from the previous experiment by pre-plotting 180s of various lengths. Alas, this failed to achieve the previous impressive result. The Cobalts won with a single loss on their side and another fairly damaged. The details of the Gardas' turns relative to the Cobalts' and the precise heading to the rear of the Cobalts' formation are actually quite important; the margin for error is pretty small. By barely microed Cobalts here I mean I basically just assigned a pair of Cobalts to shoot at each Garda and then left them mostly to their own maneuvering.
3) Well microed Gardas vs. well microed Cobalts. I actually got some ideas of how to foil Gardas' best maneuvers by observing how the less well executed versions thereof turned out in the previous test. So for this third test, the Gardas were microed as in test 1, but the Cobalts did a few things differently. First, they spread out, so they couldn't all be blindsided at the same time. I only did this in pairs initially, reasoning that if I'm going to see a difference, I should see it even at this level of effort. But after the dogfight went on for a while, the Cobalts ended up spread a bit more randomly without me planning that. The second thing I did differently was to not let the Cobalts get shot up "for free", meaning that once they we re blindsided, they didn't sit there, but bolted straight away from the Gardas to minimize the damage they took from Gardas' rear guns, while they couldn't shoot back. The third thing I did was to exploit Cobalt's superior speed to deny Garadas finalization/kill of the most damaged Cobalt. I didn't remove the most damaged Cobalt completely from the fight, but used it sparingly when its health was low, doing maybe 30%-50% shooting compared to the ones in better health. The result of all this work was that the Cobalts won with zero losses!

An interesting thing that happened in all three tests was that the Gardas basically pulled off the nearly ideal firing strategy (shooting preferentially at one target but only when its mitigation was around 20% or less) without needing practically any targeting orders! In contrast, the Cobalts needed plenty of orders so that they wouldn't raise the mitigation too high on any given Garda, especially in tests 1 and 3, in which they were given movement orders aplenty.

I've uploaded recordings of the above experiments to http://www50.zippyshare.com/v/27736806/file.html.

Now the tough question is whether the lessons learned from the well microed Cobalts in the small scale can be used in larger battles vs. Gardas to counter their rather more straightforward maneuvers. For me to even attempt this I need to find a simplified version of (3) that gets most that result. (I have some ideas, but they'll probably have to wait for the weekend.)