To me it simply looks like Apple buys out the first year of every new TSMC node and that is the main reason why the M series is more efficient. Strix Halo (N4P) has, according to Wikipedia, a transistor density about 140 MTr/mm2, while the M4 (N3E) has about 210 MTr/mm2. Isn't the process node alone enough to explain the difference? (+ software optimizations in MacOS of course)