I think it probably could, if for example the population of athletes was growing the entire time: then with a larger population, the #1 athlete (the person farthest out on the tails) will keep 'setting records'. If you draw from a distribution like the normal distribution 1000 times, the single most extreme datapoint will (almost always) be much more extreme than if you sample just 10 times.

To check this, you'd try to figure out some measure of population of Track & Field athletes (or maybe a crude proxy like entire USA or global population?), apply extreme-value theory to estimate how much the record should increase each year, and see if it undershoots the observed increases.