Usenet was tiny. Probably ~50k users, maybe 500k, back in the late 1980s / early 1990s. (Email discussion with Gene "Spaff" Spafford, one of the Usenet Cabal (TINC) members.)
I remember Slashdot ... before there were userIDs. I ... had a pretty low-digit-count ID, and remember when the site hit 100k and 1 million users.
Today, an online service with 5 million to 10 million active users is ... considered a failure. (I was ... kind of responsible for giving a good count of Google+ public user activity.) Facebook is over 2 billion users, and something like 500m regular users.
Those are big networks. And scale matters.
This isn't just a matter of the Eternal September, though acculturation is certainly a big part. I don't think Usenet could have sustained a network of this size, even with four weeks of mandatory on-boarding every September. The scale is simply too large.
One of my hobbies has been to look at how dynamics within various groups change with group size. That matters much more than the group composition (though that also matters). No matter what, a group of 5 people, or 20, or 100, or 1,000, or 10,000, has a fairly distinct feel.
Even before I'd estimated G+'s size, I had a fairly decent intuitive sense (generous, as it turned out), for the userbase size (I'd thought perhaps 20-50 million).
Reddit's various subs also give good examples of how community varies from sizes of ~1 to about 15 million (the largest subs).
My suspicion is that ~50 - 500 users is probably a relatively good sweet point, though I'm not sure how that an be utilised to make scalable discussions possible -- collecting the best from many groups, but keeping the discussions generally smaller.