There are about 20 public exchange points.
There are about 80 private exchange points.
Most exchange points have less than 10 participants.
There are a few NAPs have more than 100 participants.
In areas outside the US, the small exchanges seem to work very well. Especially where there are political and regulatory issues that might limit connectivity. It also works well in areas of the US that are more sparse concentrations of providers.
NOTE:Bill Manning's comments look like this.
What about the media? Does that make a difference in effectiveness? Many small exchange points are built out of Ethernet and that is very cost effective. For the large exchange points, Ethernet is not effective.
John Curran asks how many of the exchanges only have two parties? Bill says that there a large number of these and the number is probably growing, but there is also a number of these that die off, so the overall total number may be staying about the same.
Is anyone doing work to determine how packets flow? Yes, see the cflow work done by Daniel McRobb and John Hawkinson.
When is "critical" mass reached at an exchange? I don't know. Even when the technology broken, there are people who want to connect anyway.