Thanks for the feedback everyone. I'm glad that nobody found a glaring flaw in my methodology. I've become satisfied that if the 2003 course is correct, the race as run this year is short.
I see no reason why a major triathlon should have a less accurate run course than a USATF certified running race.
The swim is tougher. At least it was before handheld GPS and Google Earth. It is probably difficult and expensive to get an exact swim course measurement. But I don't think triathletes expect that. But with the advent of Google Earth and handheld GPS, it's a simple matter to plot the course, record the turn points, then drop a buoy at those spots. Of course, buoys can drift, GPS isn't exactly accurate, and Google Earth may not exactly line up with the depicted coordinates. But it should be possible to be within 50m or so, which is 40 seconds to a minute for most swimmers. Swim courses are routinely off by much more than that. Legend has it that the US Navy once laid out an Ironman swim course and made it 2.4 nautical miles instead of 2.4 statutory miles.
Bike: I think expectations on the bike are lower. It can't (or shouldn't be done) based on USAFT standards because in triathlon, we're supposed to ride on the right side of the road, so the shortest possible course standards wouldn't apply. But there's no reason the general principles couldn't be followed with a calibration course, etc. Even this is too much, though. I don't think most triathletes would object to the use of Mapmyride or a GPS to plot the bike course. And most people prefer an interesting course to exactitude. For example, the Ironman Wisconsin course is lollipop shaped, and the typical route from the lake to the loop part was under construction this year. The alternate route was a mile or so longer, but there was no real alternative without totally redoing the course. There aren't any out-and-backs to massage the distance, and the start/finish are fixed by the transition locations.
Run: I think it's a given that we often run more than the distance when the transitions are counted. For example, at Lake Placid, it's about .25 miles from the lake to where the bikes are racked. But there's no way to avoid that. At Kona, all athletes run a loop around the entire pier in both transitions. I've heard reports that when this is added to the run course, the total is 26.2. Most transitions are laid out so that everyone runs the same distrance regardless of where their bikes are racked. The run course should be the exact certified distance and start at the timing mats at the exit from T2 and end at the finish line.