Apple's LLVM 3.1 Clanging On Intel Sandy Bridge
Beginning with the VP8 encoding benchmark, the result was quite similar between compilers. LLVM-Clang was slightly slower than GCC 4.6.3/4.7.0, but at least the 3.1 SVN numbers are improved over the 3.0 release.
LLVM/Clang still does not support OpenMP, so the results remain a great deal behind the GNU Compiler Collection. However, at least in some of the tests there are some notable performance improvements for GraphicsMagick with LLVM/Clang 3.1 over last December's 3.0 release.