Grumpy is quite right to point out the vast lost opportunity in not having the scholarly literature freely available, but the price of this is much harder to quantify.
In response to Josef Anvil and Chris in Utah, it's worth pointing out that in many cases academics are doing this in institutional repositories--so-called "Green Open Access" in Peter Suber's jargon.
So the 5.5 billion inefficiency is the price currently paid for orienting a paper in relation to its peers, and in ordering the academic rat race: papers in journals with high impact factors being more valuable for career advancement than merely adding to the so-called "grey literature."
It's changing--20% of papers are now published in Open Access form, but it's a behemoth, and frustratingly slow to turn around, considering that the web has now been around for 20 years, and was in fact invented for this very purpose.
Over at the Imaginary Journal of Poetic Economics Heather Morrison estimates that the global cost of this system is $5.5 billion. She compares the total revenue from scholarly journals ($8 billion) with the $2.5 billion it would cost to publish every one of the world's estimated 1.5 million scholarly articles using open access publisher BioMedCentral.
Hers is a good blog to follow if you're interested in the issue, along with Peter Suber's encyclopaedic outpourings at the Sparc Open Access Newsletter.
Matt Blaze's criticisms of such "learned"* societies is entirely correct, but in continuing to publish with such regressive publications he is sadly, still part of the problem.
He pleads that his students' careers may suffer if he refuses to publish with the paywallers, but ignores the millions of scholars worldwide who would benefit from the free dissemination of information. I don't know which open access journals would be most appropriate in his specialty field, but almost certainly they exist, and he should back his cogent criticisms by actually publishing in them, thus setting a good example for his students.
*"learned" necessarily apostrophised because of the absence of evidence of learning