I'm looking at the thread safety test results across platforms and I'm puzzled by the Windows data. On other platforms, the 22.locale.num.put.mt test takes up to three minutes (at which point it times out because we don't let tests run longer than that) to either complete or crash. But in some of the Windows builds (e.g., the one below) the test succeeds after less than 3 seconds.
What do you guys make of this? Martin http://people.apache.org/~sebor/stdcxx/results/win_vista-0-em64t-msvc-8.0-12D-win32-569205-log.gz.txt