That sounds like a good idea, if I understood correctly what you propose is to start building up a Boost.Sorting library?
Yes, or Boost.Algorithm.Sorting.
Even more good news, I woke with a lot of ideas to improve my existing implementation, this is related to the way leonardo numbers are computed, the current way of doing is clearly suboptimal.
Considering that your algorithm is still in the tuning stages, and mine is in final regression testing, I suspect I'd be better off proposing a boost/algorithm/sorting with my algorithms, and then include smoothsort (and spreadsort hybrids with smoothsort if appropriate) when you've finished with it. I'd also be happy to take a look at your algorithm when you've finished optimizing it to see if I can tweak it a little. I've been amazed by the net improvement I've made with one 10 percent improvement and a bunch of 1 to 3 percent improvements over the last few months. They add up.