http://home.pipeline.com/~hbaker1/ForthStack.html (the relationship between "linear logic" (where a value can only be used once) and stack machines; Baker wrote several papers about "linear" programming; this stuff never really went anywhere, but it's interesting to think about)
I posted this because I was reading a quote of Paul Grahm, essentially saying that HLL's require less skill to implement your ideas in.
I was reminded of all of the research and work I did trying to learn to write lisp code that runs reasonably fast.
The point being that writing fast code in a HLL like lisp requires knowledge about the different techniques that are being used by your compiler to make your code efficient.
Anyway, this archive is definitely one of the better endpoints that I ran into on that search.
My personal favorites are Cheney on the MTA and the paper about a lisp architecture based entirely on destructive operations.
http://home.pipeline.com/~hbaker1/NoMotionGC.html (realtime garbage collection)
http://home.pipeline.com/~hbaker1/Prag-Parse.html (META parsing)
http://home.pipeline.com/~hbaker1/LazyAlloc.html ("CONS should not cons its arguments", a way of doing automatic memory management without giving up most of the benefits of stack allocation)
http://home.pipeline.com/~hbaker1/CheneyMTA.html (compiling Scheme into weird C whose function calls never return)
http://home.pipeline.com/~hbaker1/Subtypep.html (efficient type inference using bit vectors; see also TInference.html)
http://home.pipeline.com/~hbaker1/ForthStack.html (the relationship between "linear logic" (where a value can only be used once) and stack machines; Baker wrote several papers about "linear" programming; this stuff never really went anywhere, but it's interesting to think about)
And, not at all a research paper, but I thought it was very funny: http://home.pipeline.com/~hbaker1/sigplannotices/gigo-1997-0... (snarky satirical criticism of Ada).