This is always the paradox, there are many occasions where simplifying it for one subset of users may directly implicate people who suffer another condition. Which is partly why I keep telling people that trying to cater for specific illnesses is the wrong way to deal with problems. If you go to the doctor you tell them your symptoms and they form an idea of what you're suffering thereby treating the cause of the problem rather than the symptoms. Conditions like blindness and deafness are (in fact) on the web just passive terms to refer to people who suffer specific conditions of varying severity, making your website accessible to a blind person doesn't make your website work for someone with color-blindness. If you however look at the condition of vision loss and then target the lack of visual input you can better deal with the problem by treating it as a condition with varying levels of chronic description or severity (thereby pro-actively treating the issue, not the condition). My biggest criticism with the likes of WCAG are they act as best practices for a small number of specific issues and totally ignore the wider aspect of accessibility.
It's neither practical or possible to make something accessible for every last person on earth simply because we don't know every illness everyone suffers (like doctors, new discoveries relating to medical conditions are made all the time), that and as developers and designers we have a very restrictive amount of control over how our visitors consume what we offer. I like to believe that we can pro-actively enhance our websites to be better aided for people with disabilities (like having hand rails or wheelchair ramps) but we can't account for the hardware at the client side and thus this limited control restricts how accessible we can make content.
I agree that accessibility is part the charge of the user and the developer (in equal amounts), we can do as much as we can to make things easier for them (like for example not using PX text measurements) and by taking the time to make things a bit more flexible we thereby hand it over to the users who can adapt their method of browsing to successfully access what we provide. After all, we can't replicate the power and durability of a screen reader on the page!
As I said earlier, we cannot be held responsible for what we have no control over, we can in-fact make a website reasonably accessible but our power is limited. Accessibility is much more than what the website actually can do, as web professionals we work in co-ordination with the browser makers (who provide the medium of browsing), accessibility tool producers (who provide the tools that we ensure our stuff works within) and even to a bigger extent the hardware makers of PC's who deal with issues like speed of access, it's a pretty big undertaking but so far we do pretty well.
Some of WCAG AAA is contradictory (even in 2.0), I don't think total AAA compliance is technically possible in 90% of cases (Joe Clarke would probably agree with me on that) however I've never been much of a fan of WCAG as it's far too discriminatory in itself as to what comprises reasonable accessibility (due to it's focus on solutions rather than the issues itself). My own website is WCAG AA compliant and that's good enough for me.
Automated testing is not a solution, it's just a tool (and a very poor one at that considering how little those products can test against).
RIA's fall under the ideal of progressive enhancement, of course there will be some features which cannot be made accessible but if the base level of compatibility allows for everyone to make use of the medium (even if it's more restricted) then you can in-fact make any complex web application accessible. I hate when people refer to the law as justification for abandoning accessibility on the basis that they deem it impossible to work along with. The laws as they stand are VERY acceptable and flexible enough that only the worse offenders need be followed up and sued... in fact I'm a real believer that accessibility law is the barrier for entry our industry has been looking for to help get rid of poor practitioners, which is why I'm in favour of the EU putting more centralised laws to help boost the armoury.
The laws as they stand don't apply to personal sites but commercial entities or those providing a non-personal public service, the nature that accessibility is such a reasonable expectation means that using this law I believe we can do a lot to clean up the web and weed out the crooks / people who can't do their job properly. Disabled people don't expect or want to be able to have the exact same level of experience, they have come to understand that being totally blind doesn't mean that another implementation can give their sight back, but what they do expect is for an effort to be made to allow them to at least use it without being pushed off a cliff. Disabled people are a lot more reasonable than most people seem to assume, treating them like an annoyance isn't good.
RIA's aren't generally governed by WCAG (apart from the design element of course), you're thinking the ARIA specification which covers web apps.
My definition matches yours except in place of target-audience I prefer to use the term needs-based audience purely because the scope of target audience is too broad a definition. Where a need exists to resolve a serious conflict is where I believe that a resolution is required. Targeting your audience based on specific factors tends to reduce into questionable definitions of what accessibility is and a lot of stereotyping.
Actually the iPhone is a perfect example, it doesn't support Flash and that presents a serious accessibility issue for those who cannot get onto a desktop computer - thus it has a very high penetration if there's no alternatives available.
Actually, web based applications tend to be a lot more accessibility friendly than desktop ones because in our industry we have standards and protocols and expectations to consider these types of people, with software makers accessibility is a little thought-of expectation, in fact when I started in software development, it wasn't even a consideration. I think decentralisation of software to the web will have a serious implicit enhancement for the future of people with disabilities as we can resolve the problems on the fly with little fear that the end user will suffer issues requiring an additional layer of tweaking for needs based users.
I wouldn't be so critical of vector based formats, SVG for example is a LOT more accessibility friendly than an image (alike PNG) as you can contain alternative information directly into the file which accessibility agents can read and contextually work with.
I use a medical model of design based on my experiences working in the health and social care sector, every condition is unique and every person suffers a varying level of problems. We should be acting a lot more like doctors and investigatory scientists than "rule following minions". Even process driven designs has it's failings where as I believe a more scientific approach to accessibility not only better equips you (as your required to pro-actively seek out these issues and learn about how people are affected) but it also resolves the issue of shoe-horning and stereotyping your audience.
Which is a good thing, humans rights are valuable, especially on the Internet which houses a great deal of the worlds collective knowledge.