Not taking surge pricing into account probably plays a large role in explaining this article's findings.

Another thing to note in his section with self driving cars: the pool of cars would be fixed in that case, but the amount of uber drivers fluctuates throughout the day. Now, uber doesn't really care if drivers are 70% utilized or 50% utilized, as their revenue doesn't depend on that, but they do have an interest in keeping the drivers marginally satisfied

Also I don't have the numbers but at least in SF, taking an uber to the airport is much cheaper than taking a cab ($25 vs 50 from my office, last I checked), so I could imagine that something like that skews the distribution of ride lengths.