The crucial issue is costs. Princeton University in the United States is paying around $20 million per year for its libraries, but probably well under $0.1 million for its connection to the Internet.
When things get cheap enough, they get absorbed into the overheads. Does your department charge you for every pencil you take from the office? Back two decades ago, when I was graduating, it was common for universities to have strict accounting of long-distance calls. A professor had to call a special university operator, tell what number to call, and also specify the grant number to which the cost was to be charged. Nowadays, with telephone call costs lower, charges up to some limit are typically absorbed into the general overheads.
The title of Harnad's response to Fuller, "There's plenty of room in cyberspace," is the right response. Fuller simply "does not get it". What is driving the transformation of scholarly publishing, as well as most of the other changes sweeping the world, is the rapid increase in computing and communication capability.
I don't think we can put much hope in learned societies subsidising publications out of membership fees. The arithmetic just does not work out. To consider just mathematics, the present system of research journals costs around $200 million per year (in subscriber charges alone, which is what is relevant here). Even if we cut publishing costs by 90 per cent, we still have to raise $20 million per year.
email@example.com://xxx.lanl.gov/pg.html Not sure I understand - "subsidise" as opposed to "support" would only mean partial, and you seem to be leaving out the possibility of author page charges. Can you re-do the calculation to determine what page charges would be necessary, and then based on the average number of pages per researcher the likely amount each researcher would have to pay?
I was not trying to exclude the author page charges. They are a separate approach, and seem much more feasible. If we do cut costs by 90 per cent, then we need to collect only about $400 per paper, or $1,000 per typical publishing author per year, which does not seem so bad. Further, that $400 per paper is about what it used to cost to have the paper typed by departmental technical typists.
To get back to membership fees, even partial support would not get us far. At a rough guess, total membership fees of all mathematical societies might be $6 million per year, which is only 30 per cent of the $20 million we need. Further, that $6 million is just about the cost of Math. Rev. alone (and I doubt one can reduce the costs of MR too much without seriously compromising its quality).
Why does one way of slicing costs seem reasonable and another one not? It seems to be related to scholars' willingness to pay, and that willingness depends on whether it is their own money that is being spent. When I talk to people, it seems that they are willing to pay a couple of hundred dollars a year for memberships (for US mathematicians, it usally means AMS and perhaps MAA or SIAM in addition) and one or two inexpensive personal subscriptions in their specialty.
Personally I spend closer to $2K per year, but I am atypical, both in the number of societies I am a member of, and the number of journals I subscribe to. Also, people who run professional societies speak often of members' resistance to higher membership fees.
Scholars' institutions are willing to spend much more. A top research university is spending $5-10K per year for each mathematics faculty member for the math library. The conclusion is that we need to get institutions to pay for publications.