It's less special and more pretty much arbitrary. It's a slightly fudged mean[0] of the number of transitions matching the ephemeral second according to Markowitz, Hall, Essen and Parry ("Frequency of Cesium in terms of ephemeris time", 1957), with the original paper adopting "a probable error of ±20".

[0] of 4 measurements (761, 767, 771, 780) — the exact mean being 769.75