The kind of relative cost dynamics described in this thread leave a measurable geographic imprint on the Internet. Big network operators make deployment decisions explicitly to optimize capex/opex over a relatively short horizon, with proximate peering opportunities often justifying higher upfront costs. Conversely, there are plenty of places where lack of public IX facilities, and/or exploitive metro/regional infrastructure costs make remote interconnection not-economically-viable -- so very little peering or multihoming in general. Regions or countries fitting the latter description typically have very few autonomous networks, because there's really very little be gained from running your own network. Infrastructure (layer2, "basic telecom," etc.) was once highly regulated everywhere, and didn't/doesn't really become affordable anywhere unless/until someone in authority compels sharing or underwrites the development of competing infrastructures. I don't think it was just a coincidence that EGP was developed during the same period that Ma Bell was being broken up into regional and national "independent operating entities"...
Voila: The origin and evolving structure of IDR is a product of layer 8/9.
There's a time dimension to this dynamic as well, as infrastructure savings belatedly give rise to reduced transit costs; once and future operators jump into and out of the game at different points in the cycle. Anyone else notice how many "content providers" are now suddenly looking for peering coordinators? It's not because they expect other operators to come to their isolated data center(s)...they are building out, because that's what makes sense for them at this point in the cycle.
Now I will go back to hunkering down until SF.