Here's a graph that compares the accuracy of the GPS from an Android phone to the USB GPS dongle that I have been testing:
The bars show the (absolute) difference between the results of the GPS units to the results from the "Gold Standard" GW-60 GPS in the six GPS Team Challenge disciplines over 10 different windsurfing sessions. The unit is knots. Shorter bars are better.
The results show that the phone GPS (red bars) can give quite accurate results, but sometimes does not. In 5 of the 10 sessions, at least one number (usually for 2 second average speed) was off by 0.3 knots or more. In three sessions, the observed difference was larger than 0.5 knots, which is definitely unacceptable for competition (but still good enough for just recording your sessions, and getting an idea how fast you were).
In contrast, the USB dongle had a maximum error below 0.4 knots in all 10 sessions, with a maximum deviation of 0.2 knots in 9 out of 10 sessions. The one session with a higher error was the one I reported about previously, where the arm band with the phone and dongle had slipped for large parts of the session. In all other session, the accuracy was very good. For comparison, the observed differences when wearing two GW-60 watches (one on each hand) are often in the 0.1-0.2 knot range.
Compared to the phone GPS, the USB dongle has another advantage (in addition to the higher accuracy): the u-blox GPS chip in it can provide accuracy estimates, which allows the automatic detection and elimination of artifacts, for example those that can happen during crashes or when swimming.
Anyone interested more detail can look at the data in this PDF file. You may notice that not all sessions have entries for all 6 GPS disciplines. There are various reasons for this: some sessions are shorter than 1 hour, so GPSResults did not give 1 h results; other sessions are different length in the different files (for various reasons, including a little bug that stopped the dongle logging in earlier versions), so distance differences where meaningless; and so on.
If you're really interested and want to look at the raw GPS data yourself, you can download them from here. It's a 32 MB ZIP archive with about 35 files. Note that you may need to define the time range in GPSResults for some of the sessions where the different units recorded different length sessions if you want to get meaningful results. Also, a few of the GW-60 session contain data from multiple sessions, so make sure to select just the session you want to compare.