CSS 2.1 says the background of the root element is painted on the canvas, and does not have an exception for when the root is display:none. Since Hixie wrote a test for this, it's in the CSS 2.1 test suite. http://test.csswg.org/suites/css2.1/20101027/html4/root-box-003.htm I think it's pretty silly, but since I'm trying to get bugs filed for all our failures in the test suite and this is the last one, I'm filing it.
I think I specifically made us have the current behaviour and wrote a test for it that is in our test suite. Probably not too hard to fix.
For what it's worth, webkit gets this wrong too. I guess Opera and IE9 pass this test? Is it worth trying to make the spec saner, or not so much?
The working group changed this in http://lists.w3.org/Archives/Public/www-style/2014Jul/0162.html and also in the telecon happening right now.