With x264 H.264 video encoding, GCC was marginally faster than DragonEgg and Clang on all of the systems.
Where Clang (and DragonEgg) seem to do universally better than GCC is when it comes to measuring the compiling time for the software it builds. Clang and DragonEgg are much faster at compiling than GCC.
In terms of which compiler produces the fastest C / C++ binaries, there is no one dominant compiler. With some applications, using LLVM with the Clang compiler front-end is the fastest where as with others the mature GCC compiler holds its ground, and in others, using GCC but with LLVM's code generator and optimizer via DragonEgg provides a unique advantage. It really depends upon your particular environment and other factors (if doing any compiler tuning, debugging possibilities with different compilers, and other distinct features to GCC and LLVM) as to which compiler is superior.
From the tests today and past benchmarks, LLVM/Clang does appear to almost always compile C / C++ code faster than using GCC, but at the same time, there's still software out there that doesn't build under Clang or is problematic. There are also other compilers out there like Open64, PCC, TCC, and ICC.
Via 1103264-IV-1103253IV95 these results can be further analyzed, exported, and compared against using OpenBenchmarking.org and the Phoronix Test Suite.