I believe you are conflating TDP and power draw. TDP is thermal design power, which is an imprecise rating for system integrators and manufacturers of cooling solutions and the like to tell them how hot they can expect the component to be. If soldering the IHS leads to reduced temperatures then it also should allow Intel to reduce the TDP rating.
Somewhat, but not entirely. If, say, the 8700K was lowered to a ... oh, say 75W TDP, that would make it "compatible" with coolers able to dissipate a 20W lower thermal output. Given that the power limits of the chip usually roughly match the TDP over time (spike higher, but average out), either Intel would have to lower boost clocks or keep the power limit at 95W still.
Of course, modern boost clock schemes also play into this, where pretty much every CPU on the market boosts significantly higher than its "base" clock with a sufficient cooler, leaving the base clock as a fallback for barely good enough coolers. This tells us that most CPUs today are regulated by power limits, and not thermal limits - they clock as high as they can within the power budget, with base clocks using less power as a fallback.
In desktop usage and normal bursty workloads, lowering the TDP wouldn't be an issue, but for any situation where heat soak happens, it would happen far more quickly with a 75W-rated cooler (although solder would alleviate this again due to its more efficient heat transfer), and ultimately CPU temperatures would end up at the same point - with thermal throttling again happening, only this time to an even lower level as the lower-specced cooler now struggles to dissipate 95W of heat output. The thing is that pretty much nobody uses coolers specced this low in scenarios like this - heck, even a 212 Evo can handle 125+W - and sustained 100% CPU loads are rare outside of workstations. As such, for this scenario, the difference is this: with TIM, the die would reach heat soak at Tjunction and start thermal throttling before the IHS and cooler were at the same thermal levels, while with solder, they'd be much closer together, giving you more time before heat soak and thus longer unthrottled operation. Using an overspecced cooler - which everyone does - skews this massively, and you end up with the "get lower temps while OCing" discussion we're having today, where power draw and thermal output aren't really a part of the discussion as they're ultimately rather irrelevant.
Of course, this kinda-sorta connection between TDP, power limits and cooler specs is very much on purpose - just think of the variability of scenarios that a "95W" cooler has to cool a "95W" CPU sufficiently. Here in Norway, with our cool temperatures, that would be relatively easy even with a bad cooler. In ... say, Quatar, where you can easily reach 35-40C ambient temps (not to mention factories or other industrial plants which might have even higher temperatures) - according to the specs, the cooler still has to work. As such, they overbuild the coolers, or spec them for something like "deltaT < 60C at full fan speed and 95W thermal load" - which results in wildly different end results depending on ambient temperatures, boost clocks, and so on. In Quatar, this cooler might be hitting TjMax and throttling, or at least coming close to it, while here in Norway it would either slow down the fan, or sit at a significantly lower temperature.
Tl;dr: The relation between power draw, cooler specs and TDP is complex, but ultimately you need a cooler capable of dissipating the amount of heat energy being produced (i.e. CPU power draw) over time.
