Amazing to me that this topic jumped up today; yesterday I measured a course on the C&O Canal, and I was mulling all these things over as I did the work. Here's a comparison of calibrating on the street vs. the towpath. I have a newly laid-out 300 meter course a couple blocks from home, and a course on the towpath near Carderock, Maryland that I measured in Nov. 2009 (length 1046.83 ft or 319.07 m; course # MD09010RT).
Date Time Temp Cal Course Constant* (counts/km)
12/31 3:16 pm 55 F First Street 11236.67
12/31 5:09 pm 52 F Towpath 11223.24
1/1 7:41 am 41 F First Street 11242.50
1/1 8:42 am 39 F Towpath 11221.99
1/1 11:30 am 52 F Towpath 11216.97
1/1 12:46 pm 59 F First Street 11234.17
(*Constant is before multiplying by 1.001)
Based on the calibration, I felt I had enough information to be guided by the towpath constant rather than the road constant. But not totally, for some of the reasons Mike Sandford was stating: the towpath is not exactly the same for its entire length. My cal course is on a straight, hard and kind of bumpy stretch, but there are parts that are smoother and softer (not much real mud this time).I ended up using 11240 as a working constant, this to jibe with the longest of my towpath calibration rides rather than the final average. Probably the 20 km course is a little longer than it needs to be, but not obscenely so. I had previously marked a turnaround based on an existing half marathon, which in turn was based on another half marathon, so when I found that tentative turnaround only 39 counts beyond my planned turnaround spot I felt that it all tied together pretty well (I used the previously placed point).
I agree with Pete, let's not worry about these courses being too long. I'm in favor of calibrating on like surfaces but I think this example shows it wouldn't have been disastrous to use the road calibration.