They do. The paper's point is that plants want to choose two wavelengths with different amounts of average power. This gives the plant a window to tune within t

They do. The paper's point is that plants want to choose two wavelengths with different amounts of average power. This gives the plant a window to tune within to get the desired amount of output power. If the total illumination drops, the plant compensates by tuning more towards the high power wavelength, and vice versa. But the paper makes the point that the tuning mechanism isn't perfect due to internal noise. If the power level difference between the two input wavelengths is too great, then the random fluctuation of the tuning mechanism will itself create a lot of noise in the output. So basically there's an optimal value for the difference of average power between the two wavelengths. Now, why the plant doesn't absorb the peak green wavelength and the closest wavelength whose power is exactly the optimal difference less was unclear to me. I think the idea was that the plant also wants to minimize the wavelength difference, and a given power delta can be achieved with less difference in wavelength in the steeper sloping sections of spectrum power graph.