Now Time::Piece (porting to use base DateTime) is down to 5 tests failing. 3 of those are because DateTime doesn't support isdst (yet?) and 2 because I think DateTime.pm's implementation of day_of_year is broken (at least according to Time::Piece's test suite). The test in question creates a Time::Piece object from the epoch 951827696 which is "2000-02-29T12:34:56" (note it's feb 29th), and the day_of_year for Time::Piece is supposed to return 59, whereas DateTime's API returns 61 (it's offset is different, so using the same offset that's 60).
Counting up (31 days in January, 29 in Feb) I see no way it could be 61, no matter where you start.
Still, DateTime.pm isn't even at 0.01 yet (it's 0.01_00), so at least everything else works nicely.