When running the Himeno Poisson Pressure Solver, DragonEgg produced the fastest binaries across the board. On both the Intel and AMD systems, GCC 4.5.2 with the latest LLVM/DragonEgg code was significantly faster than GCC and LLVM/Clang. LLVM/Clang was also universally slower than GCC with this test. There was not any measurable change when upgrading from GCC 4.5.2 to GCC 4.6.0. To note though is this current test profile uses -O0 as its optimization flag.
With MAFFT, a scientific application, using LLVM 2.9 was slower than GCC 4.5.2/4.6.0 on all of the systems. Using LLVM's optimization and code generation paths with GCC by DragonEgg also slowed down the performance even more than using LLVM with the Clang compiler front-end. No change was spotted between GCC 4.5.2 and 4.6.0.