That's not how this math works. 3 is not 20% of 12, and 6 is not 40% of 9. You can't get an RA proc off of an RA, so you need to count Heavy Shots, not GCDs.
No, that's not how it works at all. The first tier is shorter (17 iirc), the rest are all 66 or 67, averaging 66.66, which makes sense because Det averages 166.66 for +1% flat. DH is only not-41.66 because it'd be "strictly" worse than Det if it was.
Also, the "nonsensical" comment was about how many things SkS does that have no relation to speed. It's kind of a bad look in the sense that any new player is gonna scratch their head at it, and it's really transparent how much SE has struggled with substats.
When? It's pretty normal for SE not to mess with substats mid-expac, so in terms of evaluating what they're doing the last relevant point is the start of 4.0. At that time they nerfed Crit (lowered the base crit multiplier), made SkS more nonsense as described, and replaced Accuracy with the very boring Direct Hit.
The effort overall is in the direction of making substats as non-distinct as possible - three that increase damage with different variance, and speed that's just kind of there because it feels like it ought to be, with a bunch of asterisks applied to it to make it more like the 'just increase damage' stats. Your suggestion is to continue this and make SkS even more of a 'just increase damage' stat.
It's all good and well to want all stats to be equally desirable by any give job, but the reality is that so long as you want to have a Crit stat or a Det stat or a DH stat, then by necessity you're saying that all substats need to boil down to global damage increases.
Because anything that isn't, will either be substantively more or less desirable than your global damage stat, and that's unacceptable under your goals. Yet if all substats need to necessarily be same-y, why even have them?
That's why I think it'd be better for SE to just decide to be fine with 'bad stats', and let players have those conversations about which job uses what stats better. Even if there are known better and worse stats for a given job, stats would be at least a little more interesting than all but two dps jobs having the same exact priority because every substat just boosts global damage.
And yes, let players be disappointed by gear sometimes, because that also means they'll be excited by gear sometimes too. Just don't make unique slots like the +35 ilvl weapon bad (or maybe just don't make any weapons bad).
A heavier diminishing-returns paradigm could be interesting too, that is to say if they found a way to make Det and DH diminish more sharply and removed the Crit bonus scaling, then for most jobs it'd be optimal to balance the three. (you'd still need to figure out SkS but yknow)
---
It kind of sounds like what you reall want here are talents - that is, you want substats to represent options that can impact how a job plays, so you want SkS to be viable for everyone since asking for a proper talent system is unrealistic.
However, as you likely know, YoshiP opposes talents because he opposes player choice when it comes to how a job plays. "False choice" and all that. They'd be equally opposed to trying to make substats a real decision. So if you want talents, you might as well just talk about how they could add talents in a non-awful way, instead of trying to backdoor the matter with Skill Speed.
For instance, one thing they could do is to make talent points an additional 'currency' of a sort, i.e. make it something where you can earn them at level cap at some controlled rate (complete with catch-up etc.) for doing certain content. SE could use an extra avenue of reward, so if they did cave on adding a talent concept, I wouldn't be surprised if it's because they want to use it as a new kind of carrot like this.
Or perhaps you think talent points could be on gear, such that you now have a fifth substat (more gear differentiation) and players would need to weigh whatever value they get from talents versus the substat. I'd think that'd be too mathy for the average player who doesn't know the nuts and bolts, but maybe SE would find that sort of obfuscation to be a positive.