"Who's to say they don't already use this capability on a regular basis? As long as you're using an unauthenticated encryption scheme you will never know"
Oh, but we could know. And we should know.
This is a good example of good vs perfect and boolean cost. You are right that we currently don't know how prevalent these attacks are and if they are employed on big scale. But it is possible to know and would make a good research topic and/or a way for the community to raise the bar and know more about the state of Internet.
Yes do MITM on unauth DH is easy. But that does not say anything about the actual application traffic flowing through the connection. A successful, _undetected_ attack must also know and subvert the application traffic to keep the illusion alive.
We could do second level authentication, out-of-band authentication etc at application level to make it possible to detect the MITM. If we do this measurement on large scale and continuously we can find out who and where these attacks are in fact occuring.
IMHO one of the best network security research projects lately is the Spoiled Onions:
http://www.cs.kau.se/philwint/spoiled_onions/
This project aims at monitoring Tor for bad nodes and use out-of-band mechanisms to find things like MITM attacks. Very cool and relevant.