In the United States accessibility can not be ignored because the enables only a small portion of the public (see: The Americans with Disabilities Act). I agree that the guidelines could do a better job of explaining which technologies to support so as to enable accessibility. But I worry that some people would be led to think that disabled populations, who are in a minority, can be ignored if the site owner desires. This is rightly illegal.
It is not illegal to have a website that's not accessible. It's illegal for the government to use or mandate software that's not accessible; and thus anyone who wants to sell to the government must make their software comply. But of course, in many cases this means that people either find some exemption or workaround, or do a cursory and not very useful job of making it accessible.
I am not arguing that you should ignore accessibility, or that it's not a good thing. What I'm saying is that there's a certain group of people in web standards who put accessibility over everything else, and refuse to adopt newer, better technologies for providing accessibility because some older, broken technology was already standardized on years ago.
For instance, there's the whole fight over longdesc in HTML5. longdesc has been standardized for a while, and lots of accessibility standards require or recommend it. It's supposed to contain a link to a page containing a textual description of the image; it is intended for cases in which you can't fit a full description into the "alt" attribute. But so many authors use it incorrectly; some just put the long description right in the longdesc attribute, some had broken links, and so on, that most screen reader users either ignored it or didn't even know it existed in the first place. One reason it's so badly broken is that normal user agents don't provide any way to access a longdesc attribute, so you never see that it's broken unless you try to use a screenreader and specifically test the longdesc.
So the HTML5 authors decided to deprecate longdesc in favor of newer solutions that were more likely to provide working description, like aria-describedby, which refers to another element on the page. That element is much more likely to work as it's already there on the page, visible to people using ordinary visual user agents, unless it's explicitly hidden. If you still want to provide a link to another page, you can just point aria-describedby at a link.
But many of the accessibility crowd complained loudly at deprecating the severely broken longdesc attribute, merely because it had already been included in so many accessibility guidelines like this, not because it was actually technically the best solution.
