Good news: it works perfectly.
Bad news: the turbo-trainer-resistance function I have been using to estimate power is totally wrong.
Excellent news: my actual power numbers are more than 20% higher than the estimate.
I will do a proper session sometime at the weekend and estimate a better fit function based on that, but for now this equation is significantly closer to measured reality than the one I gave earlier:
Power = 0.02 * speed3 – 0.02 * speed2 + 8 * speed