If you look at this illustration, you can see that the blue protein's sheet is not flat but has a fan shape, with each strand turned a little clockwise from its neighbor as you move from background to foreground. The helices, following the fan of the sheet, are not parallel but slightly skew. I think of this fan shape as the natural shape for sheets, and in the recent denovo puzzles I have carefully set up my sheets in a fan shape, and skewed my helices to match. Then I run enhanced DRW for a really long time, and the foldit score function gradually flattens my sheets and parallelizes my helices. Every time. The same happens in design puzzles - I set up fanned sheets, and the score function does its best to flatten them.
The foldit program can't "see" the fan shape, only the subscores of the individual amino acids. Somehow optimizing those subscores creates a less natural overall shape. As a human I can see the shape I want to create, but the program is working against me. Why is that? Could the score function be tweaked to allow sheets to remain fan shaped?
A related question was raised by spvincent prior to the last science chat but did not get covered: http://fold.it/portal/node/2001726#comment-31713 Sometimes you get parallel structures in design puzzles because humans like things straight and tidy. But other times the humans create curved or fanned or skewed structures, and foldit straightens them anyway.