In The Story of the Human Body on p. 85, it claims that running an equal distance at two different speeds uses equal calories:
In fact, a running human's legs store and release energy so efficiently that running is only about 30 to 50 percent more costly than walking in the endurance-speed range. What's more, these springs are so effective that they make the cost of human endurance running (but not sprinting) independent of speed: it costs the same number of calories to run five miles at a pace of either 7 or 10 minutes per mile, a phenomenon many people find counterintuitive. [emphasis mine]
It provides a reference, Energy-saving mechanisms in walking and running, Journal of Experimental Biology 1991 160: 55-69 which I looked up for more information, but it didn't have any discussion about actual human speeds.
A recent source, Economy of Running: beyond the measurement of oxygen uptake. J Appl Physiol 107:1918-1922, said the caloric unit costs of higher speeds while running definitely increase:
Caloric unit cost was 1.05 +/- 0.09, 1.07 +/- 0.08, and 1.11 +/- 0.07 kcal*kg(-1)*km(-1) at the three trial speeds, respectively. There was no difference in oxygen cost with respect to speed (P = 0.657); however, caloric unit cost significantly increased with speed (P < 0.001). It was concluded that expression of running economy in terms of caloric unit cost is more sensitive to changes in speed and is a more valuable expression of running economy than oxygen uptake, even when normalized per distance traveled.
I think this paper, Optimal running speed and the evolution of hominin hunting strategies. Journal of Human Evolution. 56, 355–360, might give some clues, but I can't access it.
If different running speeds differ in caloric unit costs, they could still be per-distance caloric equivalents, due to shorter time spent running at higher speeds.
However, anecdotally, sprinting seems like it would incur a non-linear increase in caloric expenditure per distance because of how incredibly tired I get sprinting 10 x 100m as opposed to running 1000m.
Can someone provide a more definitive, source-based answer to this question?