Hi Joerg,
Because obviously when Mc Cabe specified the way to compute the Cyclomatic Complexity there were mainly some very high level entry points (in COBOL programs for instance) and no concept of closures. From a purely theoretical point of view, each lambda should be considered as a method and so should increment at least by one the CC but on the other hand in term of human readability, I think that would be a mistake to increment the CC.
If you are aware of any academical publication on this subject, please share it :)
Thanks
Freddy