I'd like to see a cumulative version of that power law plot. How many papers had a delay of at least X days.

I have been bit by several inexplicable, long delays recently. In my case, I suspected the reason was choosing stat.ml as primary and cs.lg as cross-list, which, after my time in purgatory, was flipped (without asking me, of course). My choice was completely defensible.

It's almost like peer review all over again!