Well, after lapping my HS, I've had this nagging little voice in my head telling me to do the same from the CPU. I did the job with 800 grit sandpaper. Initially, I told myself I'd just buff what's there right now just to see if it's level. After about 30 laps in one direction and 30 in the other direction I discovered I had quite a concave IHS. So I just kept at it. Two 9x11 pieces of 800 grit later paper later I was left with a darn flat layer of copper looking back at me. I finished the job and put a mild shine on it with a sheet of 1000 grit I got from the local auto parts store just for the sake of it.
Here are a few pics and the temp. results I got from lapping both my CPU and HS. I would recommend that anyone wanting the best $20 decrease in temps should consider lapping both the CPU and HS.
Hardware details: Q6600 @ 9x333 and vcore of 1.2625V in the BIOS, P5B Deluxe (vdroop modded) cooled w/ an Ultra-120 Extreme (lapped) with Scythe/s-flex SFF21F 1600RPM fan, in a P182 case:
Each temp. point represents an average of data collected over approx. 1 h time period during the 2nd pass of a 2-pass x264 encode of a 720x480 DVD source using a high quality video profile. Data points were logged by Speedfan every 3-4 seconds over this time period. The average CPU usage was >99 % on all 4 cores throughout the experiments. Also room temp was between 20-22 °C.
This is my preferred setup: 8x10 piece of glass on a flat counter top. You can see I cut the sandpaper into a thin strip (about 2-3x the width of the CPU) and attached it to the glass with some tape. The glass is in turn tapped down to the counter top to keep everything immobilized. You'll want to moisten the sandpaper with some water, then blot it until you have no pools of water. Remember, if you get water into your chip you're sunk. Then simple hold the chip and gently move it front-to-back. I don't recommend doing circles since they tend to give uneven results.
Remember, you're after a flat chip here so don't push down on it as you lap: let the weight of your hands do it without extra pressure and go slowly so you don't use uneven pressure. After about 30 laps front-to-back, I gently blotted off the chip with a moist paper towel to remove the metal particles I just sanded off, then rotated it 90 degrees and repeated 30 laps front-to-back. Then you'll want to clean off the sand paper (add more water, then blot it damp and repeat). I'd recommend changing the sand paper frequently since it's really doing the work for you. That's basically it. You can start with 400 grit or so and lap until you can't see variations in the surface of the chip (no silver color is often a good indication that you're flat), move up to 600 or 800, then finish off with 1000 or 1200. I did mine entirely with 800 and 1000, it just takes longer with finer grits. Remember, the key is FLAT, not shiny. I would recommend that you do NOT polish the chip with a metal polish since you'll leave behind a residue that will hurt your heat transfer.
You can test the flatness at any point during the lapping process by carefully placing a razor blade across the surface of he chip and looking at the area where the razor meets the chip. Now position your eye so that you're level with the chip and pointing at a light source (a lamp will do nicely). Do you see any light coming though? If so, keep at it. Another test you can do is to take a black sharpie marker and make about 9 dots in a 3x3 grid on the surface of the IHS. Lap about 5 times, rotate, and do 5 more. Now look at the dots... did they wear off evenly? If not, keep at it. You can also simply draw an "X" from corner to corner on the chip and do this as well. Again, you'll looking for even wear.
After about 5 minutes of lapping in each direction with 800 grit. You can see how the nickel plating has come off around the edges first which shows you just how concave this thing really was:
After more lapping most of the nickel plating has been removed expect in the really low areas (the camera flash fired so close to the chip makes all the scratches show up much more so than they do under normal light):
Switched to 1000 grit, here's the result:
Another angle shows the nice dull reflection, still very so slightly concave at the extreme edges, but good enough for me:
I would recommend that anyone wanting the best $20 decrease in temps should consider lapping both the CPU and HS.
Oh, I also thought I'd mention that before I lapped the chip, I had a pretty big difference in core temps when loading with prime95 or 2x orthos: up to 6 degrees C (sorry I don't have a screenshot of this). Lapping the chip REALLY evened-them-out as you can see from the coretemp numbers after the IHS and base of the heatsink were lapped (stressed using prime95 v25.3):
The table I showed above was not based on prime95 or orthos, it was based on x264.exe which is a video encoder. It is good at using all 4 cores, but not as efficient as prime95/orthos which explains the differences in temps from that table.