Well, first I believe there are separate memories for 480p/720p/1080i (I may be wrong here), or else there are just separate memories for INPUT5 and INPUT6. One way or the other, I show different service menu values when I adjust with a test pattern from DVE (480p on INPUT5) vs. when I adjust with a test pattern from my PC (DMWVE overscan test image, 1080i on INPUT6). However whatever the actual settings are, the general tweaking approach is the same.
(1) Using DVE, I went to Title 12 (Display Setup Patterns) and then Chapter 17 (1:33 overscan pattern). On my 34XBR960 the test pattern will display automatically in FULL mode (i.e. 16x9), although I can use the MODE button to get it back to 4:3 if I wanted to (but I don't). So now I'm looking at that pattern in 16x9. It has extreme edges, as well as interior rectangles and gradation markings reflecting the various percentages of overscan.
The following recipe will probably require several iterations, going back through 2170D-1, 2170D-2 and MID3 until you're totally satisfied. So don't just stop with your first try. Also, you will want to verify that you have proper alignment on all of your inputs... since it appears there are separate memories for each.
Don't forget to write down your current settings for each input before you start.
(2) Power the TV off, and then enter service mode (Display, 5, Volume +, Power). Then press 2 repeatedly until you get to the 2170D-1 group for vertical alignment. Use 5 to go back if you "overshoot" the group.
Once in 2170D-1 I used only 0 (VPOS vertical position) and 1 (VSIZ vertical size) to adjust vertically, using 1/4 buttons to go back and forth between 0 and 1, and using 3/6 buttons to adjust that item up or down.
To sense the effect of each 3/6 tweak, push the key just one unit at a time and watch what happens on the screen. It will be obvious what you're changing, and when you've reached (or exceeded) where you should probably be.
When you're satisfied that you have centered and sized vertically (using the gradations and overscan percentage rectangles as your general guide), move on.
(3) Press 2 to move into 2170D-2, where I used only 1 (HPOS horizontal position) and 2 (HSIZ horizontal size), again using 1/4 buttons to go back and forth between 1 and 2, and using 3/6 buttons to adjust that item up or down.
Again, try your best to get the best (or expected) amount of size and centering in the horizontal direction. You may want to go back to 2170D-1 and fool around some more with vertical arrangement.
Again, use 3/6 one unit at a time and watch what happens on the screen. Even though you haven't gotten to MID3 yet, you can still sense when what you currently have is optimal or not.
(4) When satisfied that you have the vertical and horizontal size and centering done acceptably (for now, anyway), press 2 repeatedly until you get to MID3.
The adjustments with 2170D-1 and 2170D-2 are conceptually the "raster" (i.e. the "canvas"), which is sort of the background upon which the actual image will then be displayed... located on your physical screen according to your VPOS, VSIZ, HPOS and HSIZ settings.
Then, the "image" dimensions and position (on top of the "canvas") is what is controlled by MID3. So in a sense, you can have a larger background than an actual image on top of that background, and the image can be moved around and resized on top of the background using the MID3 controls.
That's why you try to maximize/optimize the "background raster" with 2170D-1 and 2170D-2, in anticipation of properly sizing and placing the "image" on top of it using MID3. Obviously you want to end up with the image size and canvas size the same, which would make the image extend out to the edges of the screen just like the conceptual raster background underneath it. This maximizes the amount of image you see, wasting nothing of your 16x9 screen real estate.
I don't know why Sony decided to implement this whole thing as "layers" (a la Photoshop) but they did. So you just have to go through this 2-step process of first spreading out the canvas to fill your screen, and then spreading out the image on top of the canvas to also fill the screen. What's up to you to control is just how much overscan you impute through your settings, meaning how much of the perimiter image you are willing to lose.
Once at MID3, you use 0 (VDHP horizontal position), 1 (VDHS horizontal size), 2 (VDVE vertical position) and 3 (VDVS vertical size) to manipulate the "image" on top of the "background" (i.e. within the "background raster"). Again, use 3/6 buttons to adjust one unit at a time up or down and watch what happens on the screen. It will be obvious what you're doing and what effect you're having.
Try to adjust things so that you have maximum amount of screen real estate covered, with symmetrical alignment of the 5% overscan rectangle a bit inside the extreme perimiter of the screen... depending on your likes. I suppose you can align things so that you have essentially 0% overscan (where the extreme outside of the test pattern rectangle is uniformly visible around the edges of the screen) but that may let in some video noise depending on what you're watching. It's more likely to see that video noise when a channel broadcasts 4x3 content inside their 16x9 digital presentation (e.g. "Curb Your Enthusiasm" on HBOHD).
In other words, shooting for approximately 1-3% overscan is probably a good idea, and will keep you from screaming at the set when you see video noise. Better not to deal with that.
(5) Once you get used to what you're doing with MID3 and the image stretching, sliding, etc., you can go back to 2170D-1 and 2170D-2 to see if you can perfect things a bit more. And then of course you'll come back to MID3 and tweak again.
This is a delicate process, but you will clearly see what you're accomplishing as you use 3/6 one unit at a time in either direction. It will be very obvious that you're either helping things or hurting things.
(6) Again, I recommend doing this for each of your inputs (INPUT5 and INPUT6 in my case, as I use INPUT5 for DVD/480p and INPUT6 for HD/720p/1080i), at least to determine where the separate memories are... are they by input, or by resolution? Based on my experience there definitely are separate memories for my 1080i (PC, INPUT6) and 480p (DVD, INPUT5) efforts.
(7) Finally, although you can tweak your heart out with a test pattern, you still want to double-check what you've done with realworld content that you're familiar with. For example, the Leno show is excellent as a test because it's wonderful picture quality and essentially a stationary image very night. Also, you might want to look at other film-based high quality shows (e.g. CSI, original, Lost, Rome) to be sure that the HxW proportions of your adusted image looks correct. Images that have the wrong H or W proportions will look squashed or stretched in the horizontal or vertical directions, and people's faces are good tests of that. Going to 0% overscan on a small screen might make the images look too small, whereas going up to 3% might be just enough "enlargement" to make things look more appealing. Of course if you've used your test patterns properly the realworld test should be just fine.
Hope this helps.