Do you happen to know if there are any published studies on this? I'd find it particularly interesting how well this number (curve) correlates with Dunbar's Number.
Pinpointing the source isn't necessary, though, only knowing that the leak occurred and the approximate number of people "in on it". Even limiting this to leaks to the media for information shared at company meetings (so the number of people is equal to the number of employees) could provide interesting data, assuming a large enough sample size (and that leaks are numerous enough).