The improvements in drivers shows there is a lot of horsepower in the silicon, a significant amount of it is still left on the table. But they seem very intent on following up on their promise to keep at the drivers.
No, DX9 gains are an outlier. They switched to hybrid DXVK implementation, where it uses the DX9On12 for some scenarios probably due to compatibility issues.
Rest you can't expect more than 15-20% in average and this is going to take months if not years. So if you look at results you can see it's behind competitors quite a lot in 1080p, but catches up at 1440p and more at 4K. Because driver issues limit the CPU, it matters less at higher GPU bound resolutions. But the best case scenario is RX6700XT or maybe on a lucky day RTX 3060 Ti.
Which is way better than what they have mind you, but 90% of the people are hyped by the unrealistic expectation of the misleading claims of press like MLID.
ARC IS behind AMD/Nvidia GPU hardware-wise. Otherwise you won't need a 400mm2 chip to compete with a 270mm2 one. They are probably behind RDNA1, nevermind RDNA3 and Ada. You can see how their integrated Xe GPU is behind AMD and this is despite the massive improvements in perf/watt and perf/mm2 that Xe has made over predecessors(2x performance in same area and power).
I am not saying drivers gaining 15-20% in average is bad, it's fantastic! But to expect it'll be RTX 3070 Ti or something is ludicrous. It'll take Battlemage to look somewhat competitive and likely Celestial to get most of it. And they'll need to continually compete on pricing.