This is what I found when I started working with Boost. I did some basic math and figured that I wouldn't really gain any significant performance improvements over using my own code.
Of course, I am running this code on a very small set of data. At the most, some of the datasets will only contain perhaps 100 elements. If I were coding against a very large (and/or complicated) set of data (i.e. pattern matching across thousands of elements is a good example) then I would rethink my using Boost.