The issue is that the variation of the sphere is smaller than the one of the cylinder. Why does that matter? In the survey phase pgfplots checks what the minima and maxima are of all plots, and then comes up with a formula such that the maximal point meta corresponds to the upper end of the color spectrum and the minimal point meta to the lower end. So all ...
You can adjust the point meta. I introduce an angle \myangle that determines the direction.