I just tested it and it seems I can manage fusion and see the effect both if I switch them and if I don't. Now I'm confused. is there something special about these images in particular or has my perception been all wrong?
As far as why it looked flat rather than inside out, I can only guess that that depth info conflicted with the other depth information you get from vision (parallax, known relative object sizes, perspective lines) and my brain ignored it.
EDIT: Actually, after trying the second view (the one with hands, not the roller coaster) cross-eyed, I saw the inside-out depth. It's most noticeable looking out the window.