I suspect that it isn't just an "internet fact" (something made up and then copied so many times it starts to look more truthful because of how many times it is seen, because not all those sources can be wrong can they?).

The higher figures may well be real, but have been taken from biased sources. If the figures come from paternity testing lab results then you are automatically biasing the data towards those that are suspicious of a paternity matter due to other correlated factors. People with no such suspicions are excluded from the results because they don't take the tests, and they would pull the average figure down if they took part in a test (but why would they think to?).