I have altered the Windows tests so that they don't set a credential on one thread and then immediately access that credential from another thread. With that change, the Windows tests seem to work reliably.
I have updated the docs to warn clients that this pattern of access is known to fail on Windows, and added a feature that can be used to run those tests on Windows in case Microsoft fixes these issues in future.
I have altered the Windows tests so that they don't set a credential on one thread and then immediately access that credential from another thread. With that change, the Windows tests seem to work reliably.
I have updated the docs to warn clients that this pattern of access is known to fail on Windows, and added a feature that can be used to run those tests on Windows in case Microsoft fixes these issues in future.
This "fixes #163" in a manner of speaking.