From the standpoint of novelty for programmers, I find the math or hardware recommendation the best so far in the original lobste.rs thread. One of the more productive concepts I've run across over the years is that there's a generalized distributive law, in that hardware tends to decompose computations as a sum of products, while software tends to decompose as a product of sums.
As to specific recommendations which retain a distinctly programming focus: Algebra of Programming (Bird & de Moor, 1997), A Programming Language (Iverson, 1962), and Transaction Processing (Gray & Reuter, 1992) are books which have caused me (as someone who started in the late structured/early oo era) to think about the informatics of programs in a more general sense than just the mechanics of programming.
Anyone have something similar, from this century? (CTM I find an excellent reference, but not particularly mind-bending) Tanenbaum once did a gates-to-OS survey book that I found mind-bending in high school; I'm guessing Nand2Tetris would be this century's equivalent, but both would frankly be beginner's books.
Upon reflection; I'd guess the limited demand (and even more limited supply of authors?) for expert books means that correspondence and papers are a more likely format than book-length for encountering ideas which are both novel and good.