It was under 200 in each tournament. If you are using just Elo rating output and analyzing that data, yes, you'd need a great deal more games than what these four tournaments provided.
My analysis was that I ran the tournaments with my rating system using the Elo ratings as the initial values to start my own analysis.
I have a two-variable rating system, while Elo's is one-variable. My rating system creates a spontaneous rating that is unbiased for each player for that tournament only. Thus, the initial rating can be compared to this tournament only rating for each player. This discrepancy is how I saw how the k=40 players were more wrongly rated than the k=10 players were.
Thus, these four tournaments were enough for me to show k=40 was the problem.