One thing that I'm strongly considering attempting right now is an autonomy system that dynamically changes a nation's "resting" autonomy. So for example, a large nation that's only done a small amount of centralization may have a resting autonomy of 40%, with any province higher than that ticking down and any province lower ticking up (with all other sources of reduction removed). It'd be less hamfisted than minimum autonomy and constant upticks, but the real issue is in the implementation. It'd have to be an event that checks every province regularly and swaps local modifiers constantly, while also interfacing with some convoluted system using the basic variable math available. I don't really want to go full Dei Gratia here, but it may be the best way.
The issue with the system that I've made now is that it has many pitfalls that can abused by the player. I still think it's an improvement, and the AI-AI interactions seem fine, but the AI-player interactions would be really abusable, without a clear solution.
The performance with Ducat Development (at least on my machine) has been encouraging, though, since it also uses recurring events that check every province. After around 20 years, as the events get desynchronized, I don't notice any performance dip at speed 5, even on the 30 day interval.