Perhaps, this is the paper I'm thinking of:
http://dl.acm.org/citation.cfm?id=1150402.1150412&coll=&...
Unfortunately I no longer have a copy of it on my computer to check. This was my understanding though from the last time I looked at it:
http://alexkrupp.typepad.com/sensemaking/2007/05/triadic_imp...
I'm pretty sure you're misinterpreting a research paper that I'm familiar with. A common measure of network structure is the clustering coefficient, which is the ratio of triads to all possible triads. In real networks, the average clustering coefficient is generally orders of magnitude less than 1%.
Besides, the network is continually growing, and people generally continually meet new people. It's highly unlikely that the FB network is going to stagnate because "everyone already knows each other". At least not for another decade...