I was developing a test rig for a cmos rom back in the 70's (yes, HP made cmos roms). I was testing my test rig and as usual I tested removing the power pin. The test kept passing so I thought my rig was screwed up. It turns out that the input signals were powering up the chip through the same protection diodes mentioned in the article.
Edit: One would think that if the input test vector included an all-zeros case the chip would fail. It turns out this didn't happen because there were two chip-enables, one true and one false.