Tough question to answer, but Reed-Elsivier took a crack at it. They looked closely at the carbon costs related to one journal (the aptly enetitled Fuel)
There are two sides to the cost: production and end use. The biggest carbon impact in journal production was employee commuting and business travel. Production of the journal used about 40 tons of CO2. Delivery of the print journal has about twice the carbon impact when compared to online delivery. But, and it's a big but, the largest variable is end user behavior. Online delivery starts out with a 2 to 1 advantage over print (say 10 tons of CO2 versus 5 tons). However, if users print out the articles, the online benefit quicky disappears. The estimate of the carbon impact of "high end use" of the online version using the numbers above raises the 5 tons to 80 tons! Although, the study seems not to include any corresponding increase related to photocopying articles from print journals.
So, production 40 tons, printing and distribution 10 tons, online delivery 5 tons. Printing out the online journal up to 75 tons!
In any event, what seems clear is that the biggest variable is end user behavior and that a marked benefit for online delivery can quickly disappear. Another reason for better e-readers!