Prompted by raph's post:
I posted about this to livejournal once: Sun patented a three-line assembly program that had already appeared in a famous paper in the previous decade. There needs to be a wiki or something for collecting these cases; I helped start one up but its server is down now.
Readability of RPN
I agree that having to track through the effects of stack manipulations hurts Forth's readability. That's why when I made up a variant of Forth for idel it had no stack-manipulation words: you use local variables for everything instead. The result seems more readable than assembly code, the more obvious language type that could've been used there.