The former. It’s more important to me that players be able to make a wide variety of characters of a given class with different Attribute spreads than for those Attribute spreads to be equally viable in all circumstances/all different class styles (e.g. close-up gish wizard vs. back row wizard throwing out big AoEs).
The game’s content should always (IMO) encourage the player to play toward the strengths of the character. In many class-based RPGs, a large number of possible stat spreads give the player nothing for the points they put into them. A 2nd Ed. fighter with a high Int or Cha winds up being just a bad fighter. It’s cool for role-playing purposes, but mechanically, they aren’t good and there are very few routes to making that Int and Cha feel like good investments.
In 3.X, WotC tried solving some of this with feats and specialty classes. E.g. you can make an Int-oriented fighter with certain feats or you can make a Swashbuckler with Int-based damage bonuses. This is a way to address the problem, but it happens on the back end rather than in the stat/attributes themselves. I made a 3.5 cleric based on William of Baskerville from The Name of the Rose. High Int/Wis/Cha, low Str/Con/Dex. It was difficult to make a character on that concept that didn’t feel like he was dragging the party down and virtually impossible with just the core rules.
In Bobby Null’s recent 3.5 game, one of the players made a high Cha fighter. He was just a bad fighter. Cool character, but really a liability in combat. Outside of combat, his conversational stats were steamrolled by the “diplomancer” noble/marshal I made. At 8th level, she had +29 to Diplomacy. Setting aside that 3.5 makes it ludicrously easy to blow out the Diplomacy DC scale with a modicum of effort, there was nothing good or fun about the massive deficiencies of the other player’s high Cha fighter. You can solve it on the back end with enough splat books, but what I would like to do in PoE is see how much of it we can solve in the Attribute system itself.
I’ve never said our current system is perfect, that it will be perfect, or that perfect balance is even a goal. I think I’ve said many things to the contrary over the course of development. It’s much easier to try to make all attributes valuable in a classless system, but I don’t think that means we shouldn’t attempt it in a class-based system.
On a minor side note, if I ever get to make an Ars Magica/Darklands-style game, I would definitely make it turn-based and classless.