...knowledge of the network and capacity to predict community consensus are more correlated with attendance than with level attained.

I'm not convinced this is true. Not sure if you already saw my response to Matt's article.
Since I'm not convinced it's true, I wouldn't be in favor of the added complexity in your proposal to add ln(attendance).

However, I completely agree with you that 6 is too high. I'm in favor of lowering that to 3 or 4.