Apple's 22 hr rating for the 16" MBP is a maximum for a niche task, it's only (up to) 15 hrs of "wireless web", which is a more typical usage and would only be about 12 hrs worse.
All that configurability of the Framework takes up space, so its battery is 15% less (85 W/h compare to 100 W/h for the MBP).
The MBP has a CPU and instruction set that was optimized for low power from the beginning, compared to x86 which has 40 years of ad-hoc cruft and assumed wall power in the beginning, so it may not even be possible to implement the whole instruction set in low-power. (Intel tried, and did not succeed. Could be BigCO ineffeciencies, but could also be that it just isn't realistic.) But Intel/AMD can't switch architectures like Apple can, because they don't control the software. There's no guarantee that the buyer of a hypothetical improved instruction set Intel CPU has access to a Rosetta program (even if Intel had the imagination to do that). On top of that, Apple has been optimizing that CPU for 15 years, and is has access to the leading node.
Additionally, (presumably because of the lack of legacy cruft) Apple has space on their die for huge caches and the GPU. On-die GPU eliminates power consumption due to an additional discrete component. Large caches also help things go faster, which means the CPU can drop down to low-power mode quicker.
Since Apple owns the CPU, it can customize the CPU for its needs, and it has relentless optimized for low power consumption, even to the extent of putting in a few new instructions for the OS.
Apple owns the OS, so it can have all kinds of power-saving features that a mass-market OS like Windows cannot feasibly implement. It is not in Microsoft's interest to take advantage of every little power savings a motherboard manufacturer might add: extra complexity (= bugs and maintenance costs) with no extra revenue potential. Linux has a similar problem, and additionally there are enough problems needing attention that I expect power optimizations beyond the big ones just do not have the interest / resources. For instance, if a 5% improvement would require a large kernel / driver refactor, I suspect it's a hard sell. Plus, macOS doesn't need to support anywhere near the number of configurations that Linux does, so it probably is less effort to do. So all those 5% increases that Apple can do add up.
Then there is the aspect that Apple can tune its OS for power saving. Update Cocoa to save energy and everyone's app uses less power. I expect GTK and Qt have other more pressing problems. On top of that, I expect Wayland and especially Xorg are not designed with minimal power consumption in mind. Etc, all the way down.
That said, 2.5 hrs does seem like it could definitely be improved.