[
users.erols.com] states that the GPS III+ has an RMS error of 7.5m (with SA off), not 15m, and these particular waypoints were each averaged over 32 readings with a reported position error of c. 12' - 14' (I know, 'cos I took them) using the WGS84 geoid.
OK, if selective availability happened to be on at the time (July 9th 2006, 20:00 - 21:00 BST) then that will have increased the position error but in any case the waypoints are actually within the error margin either way when compared with Google Earth's image.
Perhaps I overstated the case when I said that "Google Earth is wrong" - better would have been to say "Don't expect Google Earth to be 100% accurate" and I stand by the point that wrapping 2D images onto any geoid will unavoidably introduce distortion.
I'd be interested to hear other people's recommendations for more accurate consumer-grade GPS receivers, since I happen to be in the market for one at the moment.