Hi Michael,
If it's me making that decision, I opt for progression, not regression, making it work for 1.58+. There may be good reasons for considering 1.53 still, breaking changes, etc, however.
I would personally certainly prefer using solely the newest versions of Boost. However, installation of our library (the Geneva library of distributed optimization algorithms -- see http://launchpad.net/geneva) already requires a fair deal of hand-work (i.e. compilation of the Geneva code, usually). I do not want force users to have to compile different Boost versions for different versions of Geneva. And Boost 1.53 is still the default version in some common Linux distributions with a large installed base. So if I have the choice of a) forcing the user to always install the newest Boost version or b) having to go through some pains myself to make older Boost versions work with our code, I opt for simplicity on the side of the user. Best Regards, Ruediger