The April 21, 1908 NYT reports:

T.P. Morrissey of the Mercury Athletic Club, New York City, [yesterday] won the twelfth annual renewal of the Boston Athletic Association marathon road race, covering the twenty five mile from Ashland to the finish mark in 2:25:43 1-5...

The weather conditions were ideal for the first time when the contentstants, 120 in number, lined up in Ashland, ready to start on their journey of a trifle more than twenty-five miles to Boston.

The 2007 winner, Robert Cheboror, Kenya finished in 2:06:23. That's a 13.3% reduction in the finishing time over the past 100 years. This, despite the fact that the race in 1908 was 4.8% shorter than this year's race. This reflects an improvement in training and equipment but also, perhaps, incentives.

This year's Boston Marathon has 25,000 entrants and the potential pay off to winning would seem greater today than 100 years ago. The Gould hypothesis would suggest that over time runners are getting better and the long-tail approach would predict the right tail of 25,000 entrants to be faster than the right tail of 120 entrants.

I wonder about sample selection, however. My immediate thought is that sample selection was a bigger issue in the past, but that is probably debatable.