debug mode performance patch

[Paolo Carlini]

Hi,
> So here is my new attempt. This time with the changelog entry cause I 
> feel like I am close to the final result :-)
Indeed, I would say you are ;) Unless Jon has further comments today, 
please just go ahead. Only two minor nits:

     - In get_safe_base_mutex, I would just use size_t for the mask.
     - I see something inconsistent in multithreaded_swap.cc vs the 
current get_safe and in any case is affected by the issue with the size 
of the long type. If you think the adjustments are straighforward 
enough, no need to wait for approval but please post the final version 
of the patch which goes in.

Thanks!
Paolo.