There's an article by Thierry Coquand, A new paradox in type theory that gives a good explanation of Reynold's argument that polymorphism is not set-theoretic. I think this goes some way towards making important ideas about the nature of the polymorphic lambda-calculus accessible purely through online references, though a lot of important references are from the 70s and 80s and would benefit from good online summaries. I'd be grateful to hear of more references. BTW mail sent to cas@janeway.inf.tu-dresden.de will get to me, I'm quite happy that my main email address has been lost from the internet, since I don't have to spend so much time deleting spam...
jameson: Were you in Thomas Streicher's group when you were at Darmstadt? Who are you working with at Boulder? Despite my reservations, Girard's book is already something of a classic: I don't think you would have any problems convincing the librarian to try to get a copy.
Postscript: The last `Communications of the ACM' issue has as it main focus an event that I think will prove to be of great importance to computer science: The ACM has rejected licensing of software engineers essentially because code construction is not a normal engineering task. Interesting also in this light is Phil Greenspun's article reflecting on management at Ars Digita. Conclusions? I don't think software will *ever* be a `normal' engineering discipline, the way that civil engineering is today. I do think that the `Wild West' nature of the disipline will not last, however, because I think that in the next <n> decades it will become feasible to construct large systems that are proven to satisfy a formal specification.