One of my colleagues in our engineering department said the fatigue loading curve for aluminum alloys never really flattens out as the number of stress cycles approaches infinity. Because aircraft are often largely aluminum alloy construction and there are some quite old ones still operating, I thought he must have been exaggerating. Otherwise there would be frequent crashes because some metal parts would reach their endurance limits.
But after a little research, it turns out aircraft parts are indeed frequently inspected for the development of surface defects then are statistically "lifed." Frequently used parts typically do not last much past 20-25 years and are replaced on a regular maintenance schedule. The older aircraft still in use have lived so long because they have not been in constant service.
Can't say I know much about failure in Sugino cold-forged cranks (my XD-600's have worked fine now for the past two months / 400 miles), but wiki has an interesting article on metal fatigue that does indeed seem to add validity to my coworker's claim:
Note the broken aluminum bicycle crank used as an example. The S-N / Wohler (or Stress vs Number of cycles) curve is what I refer to as the fatigue loading curve. As an engineer you estimate the number of stress cycles in the life of a part in coordination with the greatest stress imposed on the most stressed part of the item, repeating the process many times as you iterate through various section moduli/ cross sections. There's always a tradeoff between light weight/ flexibiity and high factor of safety strength/ rigidity. With their S-N graph for "brittle aluminum", a part whose most stressed point is at a safety factor of 4 to begin its life will theoretically have a 50-50 chance of failure in 10^7 (ten million) stress cycles.
If for instance, your drive tire's circumference were 2130 mm (my estimation for 622 x 28), and your average gearing is 2.5:1, you would get 5.325 meters (17.5 feet) per crank rotation... That would net you 302 crank revolution per mile - if you said you put 35,000 miles on those cranks, that might equate to 10,570,000 cycles. While the cranks may have started life with a 4:1 factor of safety, the fracture at the end probably was a product of cyclic fatigue that expanded the defect(s) enough to lower it to 1:1.
dabac's point about scratches being stress risers is germane to the explanation as well, because the stress riser often becomes the most stressed part of a structure, and as they illustrate with the broken crank, potentially the original source in the ultimate failure's timeline.