I was looking into the actual implementation of thread_specific_ptr and saw that there are actually two implementations, a generic one that uses the pthread facilities for TLS and the other one some VC-based one (I'm sorry, I'm not that proficient with MS products). I don't know if people are aware of WG21 N2659, it is the proposal for TLS (thread local storage) in C++0x http://www.open-std.org/JTC1/sc22/wg21/docs/papers/2008/n2659.htm Everyone that is using POSIX getthreadspecific and setthreadspecific (that are being currently used in thread_specific_ptr) knows that they are slower than what a compiler provides. Is there any thought of actually using any of the compiler-provided keywords to remove the need for the POSIX functions and speeding up boost::thread::get_current_thread_data()?