Speeding up v3 basic_string

[Benjamin Kosnik]

> I won't argue, but consider the length of time the "interesting" tests
> take to run.  If performance drops, I'd rather know about it because
> the numbers look worse, not because people start complaining about the
> automated regression tester only finishing a cycle once a day.  :-)

Yeah. You're probably right. 

I don't know how to do this. Isn't gcc running Stepanov as part of 'make 
check' now? You'd think that same infrastructure could be ripped off for 
libstdc++.

-benjamin