Sorry for the blatant plug, but it might be relevant — in my M.Sc. thesis I surveyed main memory optimized index techniques and also provided some background for why traditional binary search is not very optimized for modern hardware. There's lots of illustrations and I cite some great material. :)

See here: https://www.researchgate.net/profile/Florian_Gross/publicati...

Along those lines:

* CSS Trees: Pointerless b-Trees with a layout optimized for cache lines (http://www.vldb.org/conf/1999/P7.pdf)

* Intel & Oracle's fast architecture-sensitive tree search (combines huge pages, cache line blocking, and SIMD in an optimal layout): http://www.researchgate.net/profile/Jatin_Chhugani/publicati...

* Adaptive radix trees (http://codematch.muehe.org/~leis/papers/ART.pdf)