In the Spirit regression test that's failing, the library that's being linked with, boost_thread-iw-mt-gdp-1_31.lib, looks like it might be incorrect. If I look at the the Boost.Threads regression tests on intel-win32-7.1, the library it's using is:
boost_thread-iw71-mt-gdp-1_31.lib ^^ The name without the version number in the Spirit regression test appears to be generated by auto-link.h. I believe I remember that auto-link should be turned off for regression tests by defining BOOST_ALL_NO_LIB (as it is for Boost.Threads, apparently), though I don't remember why it should be or where the library name comes from in that case. Anyone?
Sadly you have to turn it off if your using a toolset that produces a different name from the auto-link header - generally this is an Intel toolset problem, where we have both "versioned" and "unversioned" toolset names. The autolink header can do one or the other but not both (it's not psychic!), we really need some consistency here, but no one seems to be interested in systematising the toolset names :-( John.