In the image, you see that AC rotation accounts for ~50% of my delays. My rotation related delays have decreased a little as I've scheduled slightly longer turnarounds, but they're still higher than could be explained from the percentages displayed when scheduling turnarounds. Unless, that is, delays are carried forward and snowball (as I hypothesized above). Or, perhaps Sami just sits in a basement somewhere pulling random flight numbers from a barrel to delay them!
If anybody does know if the delays carry forward and snowball (or if, on the contrary, all flights are treated independently), that would be very helpful to know. In the former case, it might make sense to put a longer turnaround (even beyond that which gets the % down to below) after every fifth flight or so in a schedule to account for any delays that could have accumulated previously. In the latter case, there would never be much point putting a turnaround longer than that which gets the percentage down below 1%.