In our little critique of the experiment design here, I think we're getting a too hung up on specific variables.
It's interesting that what's being compared is the BEST time for each bike. I'm convinced that some comparison of speed to watts is what we'd want to compare, and however accurately watts are measured, they are measured to the same degree of reliability in every case, so a large sample may provide sufficient confidence. But instead of looking at single best times, it would be better to look at total performance for each bike.
What would a chart look like if you plotted points for the time and average watts for each ride on a bike and drew a line between them, then overlaid or compared the graphs for each bike? With enough points, I imagine one could get a clear picture. Plot watts on the y axis and time to complete the loop on the x axis. Draw a line between the points for each bike. Variable conditions and inaccuracies in wattage measures may make such lines very jagged, but one would expect the lines for each bike to tend to slope down to the right. The "fastest" bike would be the one where this whole line tended to be most to the left.