Magnus,
Don't take this as gospel, but if you have matched the iteration times as well you don't ever need to use optimization. I have found after lots of trial and error (lots of error) that I personally am better off having a suite of dark masters set at my typical iteration times for lights (my set includes 60, 120, 180, 300, 600, 900 and 1200 sec dark masters, the shorter ones needed for bloom correction since I don't have an antiblooming gate on my CCD). Then I never need to use the optimization routine which for me results in some strange results (no correlation warnings). Also, note, if you match the iteration times you do not need to include a bias frame since it is already built into the master dark.
Hope that helps,
Jim