Ok, I got my acct. on the FreeBSD64 auto tester box (the one were the most failures were occurring) and ran through the std.parallelism unit tests in a loop about 10,000 times.  Everything passed.  I therefore think that either there's some obscure bug in global locks and/or appending to an Appender from multiple threads while locking manually (which I removed from the unit test recently) or there's some obscure bug in some initialization code that only runs once even when I run the main unit test in a loop.  Please keep me posted if anyone has any further trouble with std.parallelism.