Sorry, I should have mentioned that, as it seems quite obvious. I didn't disable the solvability check.
Hmm... that's not good... my engine has an upper limit for the protein's height to width ratio. And when this is reached and it cannot solve any edge without exceeding this limit it just gives up instead of going further above that limit.... that's soooo stupid.... but with this protein the move number would probably be awful anyways, as it should have grown "thicker" according to the width earlier. But then this would require to look behind the edges to see, whether it should take the single edge completion move, which normally results in a better move count, or use multiple moves to complete another one to avoid getting into this situation.
Now I can only hope for a "good" choice of random boards for the final evaluation... *hope*