After several attempts at printing a usable pulley I threw a question into the forum to see if there was anything I was missing.
Either the pulley spurs are too wide (top right image in the gallery below), and so the belt teeth rides up on them, or the spurs are unstable and a mess (row 2, left), plus the spindle doesn't fit on the motor.
The suggestions and information from the forum was good - my Skeinforge settings needed more calibrating (note: use 100mm of feedstock for measuring the throughput, not 10mm), and nophead told me that the motor spindles are expected to be reamed.
So just as I start producing what looks like usable results, the hot end blocked and I broke the Z axis coupling. So I loaded up the 0.35mm hot end to see if I can get better results with that. Having a smaller nozzle brings it's own problems: you have to be more precise in your positioning (particularly with the Z axis), and the cohesion between layers seems to be trickier to maintain. After some fiddling in Skeinforge I was producing reasonable quality prints (row 2,right), and my test pulley prints started to fit the belt much easier (row 3, left, this is Greg's modified pulley). However the layer quality was still pretty messy (row 3, right). More playing in Skeinforge started to yield better results (row 4), by also reducing the speed in repsnapper when the print reached the spurs (even though feed and flow was already set to 15mm/s).