Point buy is interesting but brings up a whole lot of other problems... it also makes powergaming more problematic as cost benefit analysis instead of package deals becomes a bigger deal. Do you have a 25point magician quality and a 10 point one... are all spirits created equal... how much does it cost for each class of spells... etc. etc. etc.
On the one hand, any GM who wished could simply say "No home-brew traditions; use the examples provided, or don't be a spellcaster."
On the other hand, obviously such a system would need to be playtested and refined. It'd never be perfect - but then, neither is what we have now.
The only obvious problem i see is magic defense (as counterspelling is not always available) and defense against ranged attacks, because these two normally only roll with an attribute.
Good point.
For magic defense, that isn't a direct damage-resistance test or the like (IOW, that doesn't already have a significant built-in nonattribute DP contributor, the way armor is)? Double the die pool. And yes, I know that would weaken certain kinds of magic (i.e., Mind Control stuff). And I'm okay with that.

For defending against Ranged attacks ... add in another attribute, one not called on as often. I nominate Intuition - knowing which WAY to dodge/duck/twist/etc can be as useful as how fast you can do so.
So, an Agil3 Pistol3 shooter with a laser sight used to get 7 dice, and your Reaction 3 character used to get 3 to defend; net 4 die attacker's advantage. Statistically, 7 dice gets 2.333 successes, 3 dice gets 1, net 1.333 hits in attacker's favor.
Same shooter, same gear, same target but count in Intuition 3; shooter gets (3+6+1) 10 dice, defender gets (3+3) dice; net 4 die attacker's advantage. Statistically, 10 dice gets 3.333 successes, 6 dice gets 2, net 1.333 hits in attacker's favor.
At least at the "all 3's" hypothetical average dudes level, my gut check says "slightly more dice, but same general outcome. There'll be a bit mroe randomness in the system overall, but the overall outcomes should be about the same - we're just pushing the extreme possibilities out a couple hits' worth in either direction.
Still, you handily illustrate two things:
One, my suggestion would entail going over the whole system and finding possible conflicts/faults like those two, which is a pile of work;
Two, fixing them shouldn't be impossible, nor necessarily even be particularly difficult ... so the work is only a pile, not a whole mountain range.
