Our impression of the AI depends on a lot more than just its flying abilities. I mean, ultimately, the AI is programming code, and we can imagine that, like any other aspect of the game engine, it got better with every WC game - certainly, everything else did, and if you look at how mission (and, by extension, AI) scriptability advanced between Armada, WC3/4 and WCP/SO, it's just implausible that the AI didn't make the same progress.
I think, if the WC4 AI seems better than WCP's, it's because the two games are oriented towards different gameplay. In WC4, you hardly ever face more than four or five enemies at one time. It takes just a single missile to kill you. Obviously, then, the AI will feel extremely tough - it only needs to fire off one guided missile in order to get you. In WCP, the game is balanced completely differently. A bug can fire five missiles at you, all of them will hit you and... so what? You're still alive. Naturally, then, it feels like the AI is inferior. But ask yourself this - how often do you get hit by unguided DF missiles in WC3/4? Because the WCP AI is easily capable of this...
The same goes for Armada - its AI is extremely, extremely limited compared to WCP. ELTEE says it did a good job of maximising its ship strengths - but that's not true at all. In a one-on-one fight, all it ever did was fly around in circles, and its only strength was that Armada had crazy speeds and four-sided shields, so shooting down an Arrow or a Dralthi seemed harder than in other games. And no matter what ship you went up against, the basic flying strategy was the same - make sure you're slower than the enemy ship, and blast away at it while it circles around you trying (unsuccessfully) to hit you. You can't possibly praise the AI in a game where the trick to shooting down a heavy fighter using a light fighter is to give up the light fighter's speed advantage and fly slower than the heavy fighter.