This is an interesting thread -- I just finished a game (gigantic galaxy, Normal intelligence for all) where I built perfectionist starbases (every possible upgrade), and I think it was a mistake on my part. The AI wasn't smart enough to take advantage of it, but I think it was a mistake nonetheless.
There are three basic lines of upgrade with starbases -- since each of them requires separate investment, it's inappropriate to lump them all together too much. Yeah, a fully tricked-out base is one of the most valuable things in the game, but it's one of the most expensive, too. How much is too much to sink into a starbase? Here are the upgrade lines:
1. Mining/Production/Social/Economic upgrades. Some of these are extremely valuable, and can only be had by building a starbase -- so, this line is worth investment, and woe to the AI if it doesn't commit that. But let's suppose that while I do this, it does the same, or conquers planets while I'm building, and ends up with production capacity to match mine. Now what?
2. Starbase defense upgrades. Once you've sunk a fortune into upgrades of type 1, it's silly not to invest in these, right? After playing with them, I'm not sure. Consider the case where Dreadnoughts are up and running, but Avatar tech hasn't been discovered yet. To fully upgrade a starbase's defenses will cost 150 per constructor times 5 constructors, or 750 for a defensive bonus of +15. For that my opponent could buy two Dreadnoughts, or 2.5 Antimatter Missiles. That's enough to smash the starbase. But what if the starbase is protected by a fleet? That brings us to
3. Ship-aiding upgrades. These _can_ be worth the expense, but won't necessarily be. The statement that someone who attacks a heavily defended system needs to kill the starbase first is correct, so let's suppose that's what he tries to do. Let's imagine I have a defending system where my three starbases have all the attack and defense point upgrades, cost of (3 bases x 8 improvements x 150 bc) = 3600 bc for +36 attack, +3 defense to a fleet of N dreadnaughts providing orbital defense. My opponent pulls up just outside this sector with a fleet of N dreadnoughts plus twelve Antimatter Missiles... which cost the defender as much as my starbases. At this point I have three choices. The first is to venture out of the sector and attack the enemy, losing all the advantages provided by the starbases and planetary defenses and putting myself at a numerical disadvantage. The second is to do nothing, in which case I'll lose the starbases to the first dreadnoughts that attacks them. Of course, maybe my starbases have +15 defenses... but now, to be fair, the enemy has 7 extra missiles, and the starbases are just as doomed as before. Now, the third choice, and the only viable one, would be to pull several ships off the planets and park them on the starbases. Now the attacker in turn has two options: switch tactics and look to invading one of my planets (with their reduced orbital fleets), or keep the dreadnoughts back and missile the starbase defenders. (As starbase-to-ship defense bonuses are pretty miniscule compared to attack bonuses, the ships won't be safe from this.)
Okay, that's my analysis. Clearly the stacking starbases strategy can be pretty darn powerful, but unless I'm missing something, it's not unbalancingly so against a sufficiently clever opponent. Does the high-level AI qualify? I haven't tried it yet, but I'm not anxious to, either....
~SDC~