Coherent branching is also important on CPU to make use of your vector units, on GPU the payoff is just much higher. Furthermore you never have to do silly things like loop unrolling for vector unit optimization since you write all kernels as if they were executed scalar. Once coherent branching is not possible anymore you're usually dealing with random access over your data - even then the higher random access memory bandwidth can get you a speedup over one CPU socket, compared to dual socket you're about even.

Bottom line is you just have to think about GPUs in a slightly different way - they do have disadvantages (limited register/cache/memory per thread since you need so many of them), but things like branching aren't that big of a deal.