GCC 4.7 (and 4.8) are the fastest compilers for the MrBayes test with a notable boost past the GCC 4.6 series. The LLVM-based compilers were still slower with the slowest configuration being DragonEgg 3.0 and running through both the LLVM and GCC optimizers. Open64 and EKOPath failed to produce working binaries on Ivy Bridge for this benchmark.
The Blowfish performance was bad with LLVM's Clang since there is still no OpenMP support for the Apple-sponsored compiler... Hopefully it will finally materialize soon, but not yet for LLVM/Clang 3.1. EKOPath also did poorly for the reasons. DragonEgg was not any faster than vanilla GCC.
The VP8 libvpx results fluctuated a bit, but overall GCC still performed mighty fine.