Yeah, when I read the paper I thought the same as Paul: they're not measuring time since speciation, they're measuring time since the Mitochondrial Eve of each species. There's no way their methods let them do anything else.
Without even considering (as caffeine does) bottlenecks and selective sweeps, drift alone would create the same effect given long enough. How long would depend on population size; I have some spare time coming up, I'll do some simulations.