My general impression regarding accuracy, or repeatability of the sportrak and Legend, gained primarily from geocaching, and trying to return to a marked point are as follows. understanding of course that "real" accuracy of saved waypoints and geocaches is unknown:
Flat land, clear skies: The Sportrak seems more accurate and repeatable, provided you get a WAAS solution. The Legend appears slightly more accurate without WAAS.
Flat Land Tree cover: Again the Sportrak appears to do better if it can WAAS average. You need to give it 2 to 5 minutes of averaging though or the Legend will probably be as accurate and repeatable if not better.
Canyons, with or without trees: The legend is almost always more repeatable and accurate. If given 20 minutes of averaging or so the Sportrak results are pretty much in the same ballpark.
The above scattering of waypoints is the result of my trying to record the location in which I hid a geocache. Both receivers had been running for a couple of hours, the Sportrak receiving WAAS data, the Legend with WAAS turned off accidentally after I upgraded to 3.1 firmware. I assumed at the time it was just acting up having not turned WAAS off myself. The Location was on the side of a mountain, with approximately 2/3 of the sky unobstructed. Both receivers had a lock on 6 to 7 sats and were indicating EPE's in the 16 to 17 foot range. Waypoints 8-11 were recorded with my Legend. I would walk away, come back over the cache, wait 5 seconds, then record a waypoint. With the Sportrak, I averaged 5 minutes prior to saving WPT006 I then broke the receiver out of averaging, let it average 2 minutes, saved WPT007. I then broke the receiver out of averaging, let it start over again and waited 10 minutes prior to saving WPT008, left it averaging 10 more minutes for WPT009, broke out of averaging and allowed it to average 2 more minutes prior to WPTO10. Spread between waypoints for the Sportrak 26 feet (WPT006 to WPT010), spread between the Legend waypoints 10.4 feet (013 to 011). The coordinates I ended up posting were the ones at waypoint 011. How accurate they were I don't know, but people have been reporting good results.
Primarily what the above shows, as that you are going to have some error when saving a waypoint, and of course more error when trying to return to it. This plot was a result of taking time to get accurate data, and good reception conditions. As reception gets worse so will the spread between the waypoints. Don't expect to always return exactly to the same spot every time.