Thanks for the suggestion Michael. I will give these tests a try and report back for the benefit of others.
Actually, Ray, a thorough discussion of this topic is probably way outside the scope of this forum. From my engineering training, though, I can make a fairly good guess that this is a classic case of control system theory. What I mean is that you have to consider the camera's computer algorithms and the lens used as a feedback control system. When the sensor (phase detector) in the camera measures an error, the system moves the lens one way or the other to minimize the error. When it sees a small error, it has to stop the lens movement. In all likelyhood, it will overshoot and have to make a small reverse correction. Maybe several before it converges to a point that the error is below the acceptable threshold. I have noticed my lens make a large initial movement and then a couple small corrections before announcing lock.
I think the challenge to the camera designers is that the lens, which is an integral part of the system due to its mass and interitia, is different from model to model and between brands. The nature of the motor in my 28-135 lens might be the same as in the 70-200, but the mass of the glass would certainly not be. Thus, stopping the lens movement "on a dime" as they say, would be more challenging. Imagine accelerating your car and applying the brakes at just the right moment and with just the right pressure so that your front tires stopped within six inches of a line that was 100 yards away every time.
I have a suspicion that some of the front or rear focus observations (I will not call it a problem yet) are due to several factors and we may never come to a simple conclusion due to the multitude of variables. For instance:
1. Which side of focus the system starts from, or more precisely, ends on. Since depth of field has a different rate of change on one side of focus than the other, then the error seen by the system would be larger on one side when the lens comes to rest causing a subsequent corrective movement back. If the error is within tolerance, then the system is satistifed and WYSIWYG.
2. The multitude of lenses available. My guess is that each lens calls up a set of constants stored in the camera memory that affects how the AF system algorithm responds with THAT lens. If I were Canon, I would not waste too much time on testing third party lenses to determine their constants. This may be why some people have reported greater problems with non-Canon lenses.
3. The focal ratio of the open lens itself. For a given physical overshoot on an f 2.8 lens, the error would be larger because the DOF is shallower and the system could try to further minimize the error. Perhaps this is why I've heard it said that the AF systems do not work well with lenses above f 5.6. The optimum point may be very broad for the error tolerance the camera is programmed to accept.