There is some truth behind this notion. More complex shapes with more polygons will take more time to render thus slowing folding computation. Even when the molecule is hidden the rendering engine may still be updating object coordinates and other attributes. I haven't put it to an empirical test but I can believe this. One can't compare across WUs as they vary in complexity. Further if the current WU is far through its computation then the weighted average NS/day and frames/sec will carry more weight. The best way to test this would be to do 50% of a WU with one visual, note the speed, and switch noting the speed when the WU hits 100%.