I think the point has more to do with locally optimal builds than globally optimal builds (which is really kind of a spurious concept since different people are trying to achieve different things with their characters).

An example of a locally optimal build might be something like 'what is the best tripper I can make?'. It may not be strictly 'better' than a caster in any kind of global sense, but you can make much more solid statements about whether one tripper is 'better' than another at tripping.

One question you can ask about a system is 'how many different locally optimal builds exist?'. This is sort of like saying 'how many different kinds of characters are there?'. One can then ask whether adding a rule increases this number, decreases it, or leaves it unchanged. The simplest point buy systems would have a number of locally optimal builds equal to the number of mechanically different things you could buy up. You could be 'a Str character' or 'a Dex character' or whatever. Once things can interact, you can have interactions that create more local optima (combos), though this doesn't always happen if e.g. two things end up similar but one is better (this is usually the comment about casters in D&D - they can be many different archetypes depending how you build them, so the classes tend to hoard the local optima).

With some rules-light systems where all skills are the same except how you fluff them, mechanically all characters belong to the same local optimum (but at that point, the point isn't having different mechanics since you're playing rules light, and diversity is restored through narrative differences and non-mechanical interactions).