This is such an interesting discussion! Like several of you, I started out obsessed with the data and now...not so much. I know a century will take me 6 hours (even around here- we have hills in VA!) and a 50 miler is a 3 hour tour (thank you, Gilligan). Since most of my 10,20 and 35 mile training loops also average out at around 16.5-17.3 (absent stops for parkway crossings and the like), that's probably a good average speed for me if anyone were to ask (or care). But personally, I'm more interested in whether I can get up that hill one gear higher than last week (and over time, one can!) to judge how well my quads and hams are working together on climbs. And I've started not even paying attention to average speeds and instead just looking at the total elapsed moving time. If I can do the 10.5 loop in under 36 minutes door to door, I'm doing well (by the numbers, that averages to 17.5). The 20 miler has more high speed flat runs so door to door is an hour and 6 minutes (= 18.3). During each of those loops I might be going as fast as 34mph in one place and as slow as 8 or 9 mph in another, but on the flats it's a fairly consistent 25mph.

Note as well that not all cyclometers calculate average speed in the same way. Some take total elapsed moving time and total distance (better have the wheel setting correct!) and divide the two, but many actually 'sample' your speed every so many seconds or every tenth of a mile and do the calcs based on history since the beginning of the ride. Which is how you lose 2 tenths sitting at a light.