Build a non evil powergame party for BG1EE, SOD, and BG2EE.

ZeroxSP7

Also explain your choices. I’m always looking to improve my skill at just totally dominating this game.

squiros

i've been playing through bg1+2 since bgt came out. every playthrough i note which team member was the mvp and eventually created a party of mvps and here's my favorite party so far:
cleric: i have iwd and iwd2 in eet so having a single cleric is really powerful with turn undead. without iwd or iwd2 clerics are admittedly not great.

mage / assassin. mislead backstabs are great against mindflayers and beholders in vanilla. poisoned weapon on melf's is still great, but even without the kit it really doesn't lose much. UAI is great with tuigan or anything that has APR. of course, late game traps basically makes this character indispensable. specifically, firetooth xbow + any pantaloon ammo will be hilarious. thaco problems are solved with tenser's. with a small party, project image + protection from magic is the preferred way of handling liches. project image + feeblemind spam has trivialized many tricky endeavors. later on, mage thief also becomes a cleric. all the cleric scrolls can be equipped with UAI and then copied by clones to cast per day. this becomes much more interesting with IWD item abilities.

kensai / druid. healing and scaling damage is already amazing. however, disabling invisible mages by casting on party members is really helpful. even without the kensai kit, fighter hlas still make this class great with daggers. specific, this class is a self buffing firetooth dagger thrower. the good apr for dagger delivering kensai and strength bonus is amazing, even more so with druid buffs. summon fire elemental is an amazing tank, but many of the summons are really useful.

archer. simply equip gesen. indispensable for bg1, of course. doesn't fall off until level 40.

dwarven defender. with iwd items like shield of tsiernon and clasp of bron allow 100% resist physical damage while in stance. even without, this class is great for tanking in dead magic or wild magic areas. axes like k'logarath and unyielding make dwarven defender shine. with improved AI, some enemies will preferentially attack party members and ignore tanks, so this class naturally counters it really well. equip fire resist if you simply want to launch fireballs into aggroed enemies.

there is no pure mage. even on insane, most encounters are 2 or 3 rounds. encounters with 2 or 3 enemies means the enemies don't even have time to walk to party members, and as a result, melee characters tend to fall behind quickly. the casters should focus on AOE spells. this is because the archer will destroy things so fast that mages can only compete with killing stuff with aoe. in fact, time stop + improved alacrity is slower than this party, because of the cast time and delay. as a result, if the entire party had simply walked in and started shooting, the first abidalzim is still in flight when this party would have won. however, there's enough spell casting especially in the late game. if you want to load a contingency with horrid wilting, the mage assassin can still do so, especially with UAI. this is why both spell casters are multi, the late game scaling will be extremely apparent.

honorable mentions: paladins and rangers, all types, simply lack the proficiency and apr to contribute. their spells are too meager and take too long to cast. for example, casting aid and bless is already wasted time. even if the fight lasts 1 round after buffing, it's not better than this party would do unbuffed in 3 rounds.
since encounters end fairly quickly, dedicated mages tend to have very limited contribution even at high levels. this is somewhat mitigated by robe of vecna - as chain lightning is now online very quickly. wizard slayer simply doesn't compare to melfs acid arrow + melf's meteors + insects.
bards are equally useless. while the theory is good, the practicality isn't there. immune to fear is irrelevant with insects, enemies never cast things that disable party members. nor is +4 damage, it's not enough to turn a 2 round encounter into a 1 round encounter.
also, there was a thread about this in reddit a while ago. many optimal parties have the DD + archer + multi mage in somewhere. generally, a multi cleric is included for healing and buffing since clerics fall off hard.

Lord_Tansheron

Optimization is highly contingent on the framework.

If you play an unmodded game on difficulties below LoB, I totally agree: maximum offense, everything is snuffed out almost instantly. Bomb groups with efficient spells, what's left gets eaten by high-APR attackers.

That being said, precisely because most things are so easy it sometimes pays to optimize around the few encounters that aren't. Losing efficiency against a random goblin pack somewhere is a cheap price to pay for a better boss fight. The longer enemies last the more pronounced throughput differences become - especially when it's attacks vs. spells. Because spells have a 'reload time' and scale well with numbers, long fights against single enemies can allow attackers to pull ahead through consistency.

How much that matters and where to set the sweet spot can be difficult to tell. Partly because there's nonlinear elements to the games' progression paths, partly because it's tough to objectively weigh e.g. power in chapter 2 vs. power in chapter 6 (or whatever). To me personally for example, TorGal has always been one of the hardest fights because I go there very early, while on a fight like e.g. Irenicus I have everything I could ever want and then some (that's with mods, of course). That doesn't mean that TorGal > Irenicus holds true for everyone, though, because they might approach things differently, have different mods (or none) and so on.

Bard is another example of framework mattering. You're very correct in saying that in a world where fights last 1-2 rounds, a Skald's benefits will barely be noticeable. But in an SCS-modded LoB game, say, where everyone has ACs lower than a Minnesota winter, those song bonuses are an absolute godsend - and contribute more to overall damage output than an additional attacker ever would.

It all changes depending on what you're dealing with.