And even in CS you are mostly out of luck if it has been published by any of the usual anti-progress rackets like the ACM or IEEE or the other usual suspects (Springer , Elsevier and their ilk). Sometimes authors are allowed to publish a copy on their homepages, and sometimes they even do, but you still have to find it. Just yesterday I tried to find a description and current assesment of an algorithm from the late 60ies (arc labeling for shortest paths with turn penalties), and most references I found were north of $35 for twelve pages.
I completely agree about Springer and Elsevier, but the ACM and IEEE both offer "Digital Library" subscriptions that give you complete access to the entire archives of all of their publications for a couple of hundred bucks a year (including their membership fee.) That's a small price to pay for the quantity and quality of material available, at least in my opinion-- and I'd be ecstatic if Elsevier or Springer offered a similar option. (I've probably paid more than that to Springer this year alone just to get a half-dozen individual articles.)