UPDATE: Corrected "real world analysis of google’s webp versus jpg"

UPDATE: The author, Jacob Miller, has added the following information:

The tool used for encoding the images was gimp with the default settings. I updated the blog post with this information, as well as the resolution that each image was at (as some people thought that the smaller preview size in the blog was what I was testing at).

As for your test, were you using the full resolution image or did you shrink it down to the blog preview size like joelypolly (which was a communication issue on my part, as I didn't list the resolutions in the stats)? Here's the web directory that has all the full resolution images: http://jjcm.org:8081/webp , if you can get better results, let me know and I'll update the blog with your jpgs.

In particular, he's saying he compressed the 1920x1200 image, at that full resolution, and he got this result, at 45,592 bytes. Then he resized to the image we see on the blog, which exhibits the terrible banding issues.

So, I went back to Photoshop. I kept the full 1920x1200, set Photoshop's quality slider to zero, and saved this JPEG with 44,962 bytes (less than his).  Again, this is the full 1920x1200 resolution, but since Posterous resizes this image and doesn't let you view the full resolution, you can grab it from imgur.

This image definitely exhibits a moire pattern of blockiness.  To me, it doesn't look like his banded image, but it's certainly worse looking than the 580x363 images at 45KB.  The blog, though, represented a horribly banded image—is this really that bad?

I reopened this 1920x1200 full resolution 45KB compressed JPEG and resized to the resolution of his blog images, then saved as a 24-bit PNG, with this result:

For comparison, the original blog's representation of compressing 1920x1200 to 45KB JPEG in GIMP:

It seems to me Miller may have demonstrated that GIMP's JPEG compression may be less than optimal for high resolution images compressed to very small file sizes, but I am not at all familiar with GIMP, and don't know if it was the software, or if it was user error, that introduced the dramatic banding.

Updated with this 1920x1200 resolution example, I still feel the blog's images as presented are not representative of typical JPEG compression tools and usage.