The commercial GPS units have the same resolution as the military units, ever since they removed "selective availability". The GPS may only be accurate to x distance but the accuracy is not random, it doesn't rapidly jump all over the place from sample to sample. It does drift over time though, so you will notice if you ride in a stationary position for an hour you may get a hundred feet or so of "movement" detected. You have a far higher chance of your tire circumference being off enough to accumulate more error than with the GPS.
Maybe I'm not understanding this correctly. As I understand it, GPS cyclocomputers do not calculate a distance from point A to point B by doing a vector analysis between point A and B. If it actually does that, I would agree that if A and B were far apart, the GPS system will give a more accurate reading. Instead, these cyclocomputers tally up a series of small distances into a cumulative total distance.
However, as you said so yourself (underlined), there are fluctuations in GPS units that introduce significant errors when measuring small distances. These small errors, when added up, will give you a larger error than the tiny 0.3% error (estimated in my last post) of a conventional cyclocomputer.
Taken another way, what is the smallest distance a device can measure with any sort of accuracy? For a conventional cyclocomputer, it depends on the wheel circumference (at least for the Cateye) so about 2 meters (or less if you have smaller wheels). Every revolution of the wheel will give you 2 meters +/- 0.3%. For a GPS unit, it's significant larger than 2 meters so already you're behind in accuracy. Of course, I'm sure the GPS wizards have algorithms that can reduce this error rate, but you're still stuck with a system that's not really intended to measure small distances accurately, while being required to add up these small inaccurate distances.