FWIW, if anything at all, the vast majority of AF adjustments I have made to my multiple combinations of body and lens have been to correct front focusing. I guess on average I've dialed in around +5, but anything up to +10 is not unheard of with my gear. I did briefly own a Sigma 10mm fisheye, but that needed at least +15 on my 50D, and even then I wasn't sure about its performance. I wasn't happy with that level of error and returned the lens.
I might add that I have read (only once, I admit) that a little front focus may be designed in on purpose. If true (and I have no idea if it is), I imagine the benefits are twofold - (1) in situations where DOF extends further behind the focused point than in front, which is most times, a little front focusing could actually give you a bit more of the scene in focus than you might otherwise get; (2) for shooting oncoming action, a little front focusing might give the camera an advantage in being focused closer to the right place when the shutter is released, and you are more likely to shoot an approaching subject, typically, than the tail end of one disappearing into the distance.
When performing these tests it is important to consider the performance expectations that are reasonable for a given lens/body/AF point combination. For a lens which is slower than f/2.8 when wide open, or any lens focused with an AF point lacking a high precision sensor, the focus accuracy is only good to somewhere within the depth of field. That depth of field is the DOF when making a "normal" sized print for viewing from a "normal" viewing distance. e.g. a 10x8 print viewed from 12" away. That specification should allow you to have a sharp subject, even if only just sharp, under "normal" conditions of photography. If you are going to view output from a camera like a 5D2 at 100% then you are talking about a vast enlargement beyond 10x8. That is not "normal" photography and will reveal all sorts of problems that "normally" you would not see.
If you use a lens of f/2.8 or faster, combined with an AF point with a high precision sensor, then you should expect AF accuracy to improve to within 1/3 DOF. That's a big improvement, but one which still may not stand up to scrutiny at 100% viewing.
Now let's translate that to some real numbers, taking a 200mm f/2.8 lens as an example, focused with a high precision sensor. Canon (well, Chuck Westfall of Canon) says that AF testing should be performed at a distance not less than 50X focal length. That means testing at 10m or more, not MFD of 1.4m or so. Using DOFMaster as our calculator we get a DOF with a 5D2 and 200/2.8 lens at 10m distance of 41.6cm, extending from 20.4cm in front of the subject to 21.2cm behind. Taking 1/3 of that we basically get a DOF of +/-7cm when viewing a "normal" sized print (10x8) at a "normal" viewing distance (12").
Now, viewing at 100% is going to give you a virtual image from a 5D2 of (depending on your monitor) something around the 40" to 60" mark, or 4X-6X larger than "normal". If you want to make a judgement of whether the equipment is performing within spec then step back from the screen and view your results from 4' to 6' away, as appropriate. Does your focus target look sharp? If so the camera and lens are performing within spec. If there is something you can do with AF microadjustment to make things even better then that's a bonus, but don't be at all surprised if, in order to achieve perfection, you will have to make those adjustments. If you wanted Canon to produce kit to that level of accuracy off the production line we would all be paying dearly for the additional production costs.
The limitation with AF microadjustment is that you can only have one adjustment for any one lens, and that adjustment does have to make do for all focal lengths and focus distances. If for some reason you cannot get the results required to cover the full operational range of the lens - e.g. at 70mm it requires -6 but at 200mm it requires +12 - then that's not something you can fix. It will have to go to Canon (or Sigma or whoever) to be calibrated properly.
So, the real question is - can you make the AF performance of your body/lens good enough (not necessarily perfect, but good enough) or is it basically beyond help and needing the manufacturer to fix it? If the lens is under warranty then send it in and Canon (or Sigma etc.) will sort it for free. If you want the best hope of performance within spec then send the body in too, but don't expect perfection when the kit comes back. It should be operating within spec, maybe a little better, or even a lot better, but I don't think Canon calibrates specific bodies and lenses to each other. That would be silly, as other combinations might then not work. All they can do is to make sure each component is individually working as it should.
As for the nifty, despite the fast max aperture, its AF motor is really crude and not very accurate. Focus once and you may get a razor sharp result. Do nothing but hit the AF button once more and it may go out of focus. Repeat again and it may come back into focus, then out, etc. etc.. That is why you need to establish a pattern to the focus behaviour. A single test shot is not going to be conclusive, and you also can't guarantee that in the real world the nifty's focus will nail the shot every time. If you are doing critical shallow DOF work, such as a portrait, you may well need to take two, or three, or four, or more shots to make sure you get one in the bag. One of the reasons I moved up from the 50/1.8 to the 50/1.4 was in the hope of superior AF performance. Unfortunately the improvement is slim. Neither lens has the full blown ring USM focus motor of superior lenses. They may deliver adequate performance for "normal" viewing, but will probably leave pixel peepers permanently frustrated if the eyes aren't always razor sharp.
Pixel peeping is the curse of the manufacturers. If the pixel level noise doesn't get them it'll be the diffraction, and if not the diffraction it'll be the AF (in)accuracy, and if not the AF (in)accuracy it will be the subject blur, and if not the subject blur it will be the camera shake. For a given cost the manufacturers can do so much, but one should not forget that a lot of weight falls on the photographer's shoulders to up his/her game too, if they want to play the pixel peeping game. Once upon a time we used to be happy taking pictures. Now people seem more concerned about taking pixels, and making sure the individual pixels themselves are individually perfect. Once in a while we should stand back and see whether we have in fact got a photograph we can be proud of. The pixels are important, but not as important as the pictures, or am I wrong?
