Short version: I didn't find any extra drag wearing a backpack!
I was curious about just how much extra drag was caused by wearing a backpack, so to test that I coasted down a hill while recording the instantaneous speed at each point, and then compare at the end. I started from a dead stop, no power supplied other than gravity and (unfortunately) the breeze from an occasional car. I did the coast-down several times, both with and without the backpack which was half-filled with plastic bags to puff it out. Since this is commuting, for the tests I dressed in knee-length cargo shorts and a loose cotton polo shirt. I rode down in the drops, not completely horizontal but my back is 10 or 15 degrees from flat. There was no wind today.
Here is the comparison, with an average of the backpack runs vs average of the non-backpack runs. These are also smoothed with a 3-point moving average. Distance down the hill on the bottom plotted against speed on the y axis:
I expected some difference if even only a little, but if anything the descent was faster with a backpack than without. The difference is within the error margin though.
The graphs of the individual runs:
As you can see they are all pretty consistent, enough that I don't feel any pressing need to analyze the deviations further. I had to throw out a couple of runs due to excessive traffic, not pictured here.
Regarding the data collection, if anyone's interested, I used the magnetic reed switch from an old bicycle computer tied into a microprocessor board (Arduino Uno) which saved the time of each switch closure to a micro-sd card. I used software de-bouncing and I decided on the leading edge of the closed segment for the time stamp. These switches aren't as precise as you'd think by the way, with early or late closures seemingly at random. Hence the moving average - an early closure makes one open segment shorter but the successive one longer by the same amount, and a moving average evens that out.
It's this hill
http://goo.gl/maps/Y3sDY
The bike:
and recorder
