Those photographic comparisons are pretty subjective and zoom in on muddy/blurred areas of the image for comparison - it would have been nice to see the entire photos. The non-photographic images clearly give JPEG2000 an edge, but he then compares it to PNG. Lossy compression isn't for compressing things like pictures of text, which this guy notes but makes the comparison anyway.
The output quality is also codec dependent, and the one used in those tests has been shown to be towards the back of the pack in a 2005 shoot-out[1].
There is also no comparison between lossless JPEG2000 and PNG on a photograph.
