Facundo, first of all: *really* nice work, thanks a lot.

While I don't fully understand the issues raised lately here, 
especially what broke (user code). I can see, that it completely 
solves my original problem, which is great.

While reading the patch, I noticed, that the modifications to 
Doc/library/urllib2.rst contain two typos ("retrive" instead 
of "retrieve").

The only remaining issue left in this area is using some form of 
ordered dict for headers in order to control - how it comes - the 
order of headers ;-), but that's left for another issue to raise some 
day.