In case you’re not aware, WCAG 2.0 doesn’t insist that web pages should be valid. It instead insist that they must be able to be parsed unambigously. Basically, this means that your nesting of elements should be correct and you shouldn’t have overlapping elements. The only publicly stated reason that I could in anyway see that relates to the dropping of validity is the desire to make WCAG 2.0 technologically neutral.
It does not, and should not contradict technological neutrality to say that technologies must be used according to specification.

I have heard rumour, yapping like a barking dog, assert that validity has been pushed to the back of the queue because large companies and organisations don’t necessarily want to have to insure that their web development tools produce valid code. This may well be complete nonsense. I certainly don’t make any claims for it’s validity.

Okay, why should sites be valid?

Because things should be done according to the standards laid out for them.

I’ll rephrase that: what difference does it make to accessibility if things aren’t valid?

We don’t know. And that’s precisely the problem. At any given time, all the currently available web browsers may be able to support invalid markup. We cannot guarantee that this will always remain the case, and yet we do know that any web browsers produced according to their standards would support valid markup. Using valid markup is therefore a reasonable and way to ensure that the principle "Content should be robust enough to work with current and future user agents (including assistive technologies)" is adhered to. To make this technologically neutral, simply state that Web units or authored components are produced to the documented standard for that technology.

Netscape 4 added a layer element that was not part of the W3C standard which caused problems in some cases when you encountered a layer element with a browser that did not support it.There was also the problem that Netscape and Internet Explorer each supported their own way of marking up shortened terms – either <acronym> or <abbr> but not both. If there is no requirements for documents to be valid, there is no reason for user agents further down the line not to add new elements that only they support.

Browser wars come at the expense of accessibility. Back accessibility, back standards, back validity. Send a comment to the W3C telling them we want validity back in.