VD --
If you look at the factory-fresh TR U-120-Extreme's base, you can see that it isn't a random irregularity caused by loose tolerances or sloppy workmanship.
The base is a smooth and uniform convex shape, with a ridge line running evenly and straight across the heatsink base. It isn't a random irregularity of shoddy machine-shop work; it is a consistent feature among all the coolers produced.
Based on discussions I've had with their tech-support, it was a conscious design decision. That is, the tech-support calls it a conscious design decision, and I call it a miscalculation or flaw in design. But, as the OP here shows, it is correctible. The convex design or a perceived need for lapping may be a factor deterring someone from purchasing the product. Even so, for the cooler that has had such rave attention at this forum for months now, there was no special lapping of the unit for the tests, and it still proved superior to the others. But something should be said about the test configuration for the Ultra-120-Extreme -- and it is really quite simple.
The Anandtech tests were made using a Core 2 Duo X6800 processor. When TR designed the cooler, they expected that users would install it so that the high-point of the convex ridge would cross over both cores. If one needed to flip the cooler around 90-degrees, the base would not contact the cap optimally, so lapping would be mandatory.
With a Core-2-Quad, it would seem that lapping would also be mandatory, for the C2Q is essentially two dual-cores placed side-by-side. Now, these remarks are just my opinion, but I've seen enough from people who've had problems one way or the other with installations showing temperatures that should've been lower, and I've watched the discussions about this as the user made progress in making the cooler work as it was supposed to work.
On the matter of lapping, smoothness and polish, let me say this.
Metals of different hardness require a different grit to make a mirror finish. A hard metal, like steel, can be brought to a mirror finish with 400 wet-or-dry. Copper is a softer metal, and to bring it to a polish, a finer grit is necessary.
A mirror finish is not necessary, or it is less necessary depending on the sort of thermal interface material used. Arctic Silver 5 is tested by the manufacturer in a layer that is 0.001 inch thick. The size of the silver particles in the paste are within a micron or so, which I think is one 25-thousandth of an inch. Similarly, for nano-diamond or IC Diamond thermal paste: the diamond particles are less than 0.6 micron in size.
The particles need to make as extensive a contact as possible with the heatsink base and processor cap. But those particles will also fill in any scratches in the base. The thermal resistance of the paste is a known -- it is a given -- and superior TIMs have very low thermal resistances, so whether the heat must travel through one particle or 50 particles between the cap and the base is not significant. Either the silver or the diamond particles have significantly greater thermal conductivity than the processor cap or the copper heatsink base.
A fine polish is not necessary, and fine scratches will collect TIM particles to leave the overall surface smooth. You have to rub the base vigorously with alcohol to remove the stain of those particles when you clean it.
The significant causes for degraded heatsink performance are air-pockets and poor contact. So a flat surface with fine scratches is infinitely superior to an irregular, convex or concave surface with a high polish to it.