I think the article is OK overall even though a few points are not very well phrased.
As for image attributes: while they don't accellerate loading files, they enable the browser to render properly while images are still being loaded, so the user can start scanning the text immediately without elements bouncing back and forth.
If an image doesn’t have these tags, the browser has to wait until the image loads before it can go on loading the rest of the page.
That's simply not true at all, as is this:
By using multiple domains (even if they’re using the same IP address) you can download objects at the same time, where on a single server you’d have to load one at a time.
I'm not convinced they are wrong but the reasoning is poorly explained or misleading.
Without knowing the size the image will take up, the browser may be forced to rerender the page as the attribute data is received from the images to make way for it. This doesn't need to be done if there are attributes or if you've defined the size of its container etc.
For the second point, it's a comment that browsers have a limit to maximum persistent connections per server and a larger limit for maximum persistent connections overall. It naively uses the domain to define what a server is, so splitting over multiple subdomains will make it faster (see http://www.ajaxperformance.com/2006/12/18/circumventing-brow...). Incidentally, one of the big changes in FireFox 3 was to increase the number of persistent connections per server and IE8 is going to do the same thing. (See http://blogs.zdnet.com/Burnette/?p=565)
As for image attributes: while they don't accellerate loading files, they enable the browser to render properly while images are still being loaded, so the user can start scanning the text immediately without elements bouncing back and forth.