Alexander, this slipped off my radar.
Some of the tests Brian added fail with the patch applied. I fixed most of them by adding back the support for converting an aware datetime to an aware localtime. The remaining two I believe are what you are fixing with this patch (in addition to converting to using astimezone for the current-localtime case). So there I fixed the tests.
Hopefully I got this right, and hopefully you can find time to review it. I'll probably check it in before the RC regardless, since it looks like an improvement to me.