> It probably is _always_ the case.
That is trying to have it both ways: "probably" and "always" fight with each other. Probably so, or most often. But not always. In technical terms, this is called "weak convergence" or "convergence in probability." Here's Marx explaining the law of large numbers to the best of his lights:
"In every industry, each individual labourer, be he Peter or Paul, differs from the average labourer. These individual differences, or “errors” as they are called in mathematics, compensate one another, and vanish, whenever a certain minimum number of workmen are employed together."
Furthermore, adding the first additional worker (e.g. adding Engels to Marx) gives you the largest gain in converging to the average -- again, not always, but just in probability.
http://www.marxists.org/archive/marx/works/1867-c1/ch13.htm