I noticed the same thing as the above poster when looking at the tracklog of my ride yesterday. One of my trails is a MUP that runs on one side of a canal. The path is paved about 8-10 feet wide, I started down it around 5pm, and crossed the same point coming back around 6pm, which is plenty of time for satelite drift, ionosphere change, barometric and temperature change, and so on.

Still, looking at the track log, you could very clearly tell which line was out and which was back because I always ride on the right-side edge of the path and the GPS was consistent enough in its accuracy to determine if I was on the east or west edge of a single MUP.

GPS isn't perfect, no. I know this from geocaching that when you're sitting stationary, it can take a good 30 to 60 seconds for it to figure out exactly where within that 10-15 foot range you are, but for cycling I've been amazed with how NOT inaccurate the tracklogs seem to be. It's almost as if it has better accuracy when you're moving than when stationary. Go figure.