I've identified this as a bottleneck in my program. I know that the answer is to profile, and I've done so, but my results are inconsistent and vary with different optimization settings. Later on in the development of my program, once we've settled on which optimizations we are using, I can get a better answer, but I'm sure it also depends on the compiler in use, and maybe other factors, so I'm wondering what a good general best practice is, so I can hopefully pick the best choice now and not have to change the type signatures of a whole bunch of functions :)