The idea behind having 3 methods is that setFrozenTokens() is a super-power that explicitly “resolves” the race condition by ignoring it.
As a superpower it wouldn’t be available to most of the parties who are given the ability to freeze or unfreeze a quantum of tokens - and in the common case it is probably set not to be accessible to anyone.
I don’t understand the practical scenario where the race condition described for the delta methods arises - why are there 2 parties making overlapping demands on what should be the same tokens?
I do appreciate the gas efficiency argument - but does it hold true for a world where in practice you generally use the delta extensions?
The core question is about the use cases. If the assumption is that there is only one party able to freeze, and they do the relevant calculations offchain, then the absolute value makes sense. But that seems like a bad assumption to me. I cannot imagine a good reason to enable setting an absolute value (beyond the theoretical purity of making everyone work from first principles) unless you have one omnipotent central authority, and I can see plenty of problems with enabling that approach since in practice you probably then need to hem it in with extra checking any time it is exercised to ensure it isn’t a malicious over-reach.