It seems like integrating this kind of decision making into that spreadsheet wouldn't be that much more complicated than the current mechanics - have threat be inversely proportionate to health and have spellcasting trigger a temporary large increase in threat which dissipates after the spell has been cast. So a fighter would be at a consistent medium-high level of threat, which will generally be higher than other party members, except when a spell is being cast. Maybe have the time to react to a change in threat be longer for enemies in melee to prevent them rushing around like chickens with their heads cut off when you put your casters at opposite ends of the combat area. Certainly more convoluted than simply having enemies react to events as you describe, but not more convoluted than the way its done. And at least its something a spreadsheet monkey can work with.
Am I missing something? If not it seems possible that this lack of effective enemy target selection is a feature of the system - creating facsimile of enemies reacting tactically, so that players will have to react to a changing situation, but limiting the effectiveness of the enemy "tactics" to something that can be addressed by the player without having to make hard decisions (like having a spellcaster be dormant until most useful). Which isn't to say it can't be made difficult, just that when difficulty is increased it is done so by increasing the demands of micromanagement rather than tactical decisionmaking.
Not bad ideas. However, consider:
1) Spells in the game are mostly instant-use rather than charge-up.
2) Enemies need lots of instant-use, preferably ranged abilities to actually counter spellcasters effectively.
3) Enemies need to be smart enough to use their abilities tactically rather than get caught with their pants down by "blowing their load" too soon.
4) Battles have to last long enough for the player to actually be able to see, understand and appreciate enemy tactics.
I'm not saying it can't happen, but it's not just threat that matters. A value that tells the AI "bumrush the mage when he casts a spell" might even be a weakness in certain situations. You need to have special-case scripts that appeal to specifics.
And all of that is kind of invalidated if your combat designers have decided that all encounters need to last 30 seconds exactly so ADHD players don't get bored and/or can just hack and slash through stuff if they want to. Watch almost any Dragon Age LP and you will see exactly what BioWare have to contend with - players using their main character only instead of the entire party, players not using spells, players auto-leveling everyone, players constantly charging into glory only to die and then get frustrated when doing the same thing 5 times more doesn't work and concluding the game is too hard, etc.
So basically, making an RPG intended to sell to more than a pretty small demographic is almost necessarily going to mean you either dumb things down or piss off a lot of players who never finish your game.