I was shocked at user srkirt's discovery and photos of the massive amount of thermal compound used to fill the cavernous void between the die and heatsink, so I did some experimenting on my 2020 base i3.
First, apologies for my failure to take pics of the process. I cleaned up the die and heatsink with alcohol, the thickness of the layer of thermal paste was insane. As for the heatsink, the portion that lays directly over the die is actually milled out ever so slightly, which increases the gap between the die and the heatsink, madness. I didn't want to use a soda can, as I figured the plastic/paint layer on the outside would affect performance. I ended up using a disposable aluminum foil pan, used for serving food. I cut three layers of foil and sandwiched them together with Noctua NT-H1 thermal paste. I initially tried just two layers, but with two layers and a thin layer of paste on the die, when placing the heatsink on top it didn't fully contact the die, so I added a third layer. With three layers, contact between the heatsink and die was much better, the heatsink would rock back and forth ever so slightly when laid on the die, so with it screwed down it should be making good contact with positive pressure against the die.
Both before and after the mod, I ran three back-to-back Geekbench CPU tests, followed by a single run of Cinebench. Results speak for themselves in the attached screenshots. The Geekbench shot is taken just after the third run ended, both before and after the mod, and the Cinebench shot shows before and after, again just after the test ended. Both tests show that with better cooling, the CPU is able to run at a higher wattage, and at a higher clock speed in both sustained and burst scenarios. As for the Geekbench scores, I also ran separate sets of 3 tests both pre and post-mod, allowing time to cool off in between. The average multi-core score with stock heatsink was 2145, and with the modded heatsink it was 2309, single-core scores were nearly identical.
My triple-layer foil shim is obviously a crude hack, and having four layers of thermal paste is definitely not ideal. A proper single layer shim, ideally made of copper, and only two layers of paste would probably yield even better cooling/performance. I also think this mod would have an even bigger impact on the i5/i7 CPUs, assuming the heatsinks are of the same garbage design.
Every product design is all about tradeoffs, we in the Apple community are acutely aware of that. However, I fail to see what tradeoff was made in engineering an inadequate cooling system for these MacBook Airs. A tradeoff would be good for bad, positive for negative. Making the heatsink a millimeter thicker would have resulted in better cooling and thus better performance, that's a positive. What I fail to see is the complementary negative to that positive, because there can't possibly be one. Weight, cost, heat, noise, price, none of those would be negatively impacted. Releasing the product as-is, with an inadequate cooling system, is itself a negative, so where's the positive side of the tradeoff? Again, there can't be one.
Apple clearly made a decision with the design of this cooling system. The cynic in me thinks it was for product segmentation purposes, keeping the Air performance below the Pro, and/or keeping it below the performance of the iPad line. Someone please prove me wrong, because I don't want to believe this is the reason.