Because the decentralised ACL implementation (state resolution) relies on power levels as a simple comparator for determining merge conflict resolution. State resolution is the academically novel bit of Matrix and hard to get right, and while we can swap it out for role based ACLs in future, this is orthogonal to Spaces. Meanwhile it honestly hasn’t come up as a requirement, even from enterprisey folks, given we push power levels as hard as possible.
I should also add that there are at least two matrix spec change proposals flying around to add role based ACLs, and one of the main reasons for switching spaces to be built on rooms is that they will automatically inherit the ACL model when it lands :)
Solving conflict resolution should be doable - it's just that it's literally the most critical and important bit of the entire Matrix ecosystem, and any changes have to be done very deliberately and carefully, ideally with formal proofs that it's not going to break stuff. It's the equivalent of switching out Git's merge algorithm (https://github.com/git/git/blob/master/Documentation/technic...) or switching out for a new kernel scheduler implementation. Any bugs or thinkos there mean that the whole fabric falls apart, hence wanting to consider it separately to Spaces.
https://matrix.org/docs/guides/implementing-stateres and https://matrix.org/blog/2020/06/16/matrix-decomposition-an-i... give more info on how state res works and why you wouldn’t rush to jump to role based ACLs.
It’s nothing whatsoever to do with IRC or chanserv style behaviour, beyond coincidence.