This talk is the most interesting material I've seen in years. I first saw it some time ago and it has stayed on my mind. I believe Steele is entirely on the right track.
I highly value "Programming Pearls" by Jon Bentley in this regard. He proposes a problem and then solves it in naive way followed by the right way/ways. I realized the importance of choosing correct data structure for a particular problem from that book.
http://www.infoq.com/presentations/Thinking-Parallel-Program...