While there is little doubt that archives such as the famous physics e-print library "xxx" could replace entirely the scholarly journal as a medium of fast communication, they have not done so for two reasons other than the obvious one, inertia.
Firstly, there is the issue of peer-review. The wide range of content, especially in usenet newsgroups, is obvious to any longtime cybercitizen. While xxx, thanks to the so-called LaTeX Barrier (insistence on the use of a word-processing language well-known to scientists but evidently not to the public at large), has maintained its research orientation and academic journal-like standards, papers from people who are not experts will arrive there in years or more likely months, and the signal to noise ratio will fall to unbearable levels. Peer-review allegedly guards against this danger in journals, but its evils are too well-known to require repetition here.
Virtual libraries, if staffed by conscientious genuine scholars, could resolve this issue without invoking censorship by:
Secondly, there is the problem of locating information. In reality, this is a somewhat specious issue: locating information in scholarly journals was extremely difficult before the advent of services like Dialog and Melvyl, requiring laborious year-by-year searches of indexes which often contained only keywords. The virtual library by its structure duplicates the more advanced commercial searches, but is available free-to-all.
Hence we propose that archives in the style of xxx, but with extensive logical subject indexing, be promoted in all branches of learning as the natural next step in the evolution of journals. It is important that a mirror system and permanent serial-cataloguing of entries be maintained, as at xxx.