There's a saying attributed to Mark Twain that one should never argue with a fool because onlookers may not be able to tell the difference. I've seen a lot of Web people arguing about accessibility, design, and typographic units, and I am forced to agree with Clemens.
I'd like to see some compelling arguments. There are none. Units are units, and whether the browsers actually do the right thing or not is quite irrelevant to the question of accessibility. A site cannot be made accessible by hacking it to coincidentally work around defects in specific browsers.
Accessibility to a building is not a property of the bricks and mortar that make it up, but an emergent condition stemming from the challenges a person faces when navigating the patterns present in the structure. We cannot design buildings to address every challenge, so instead we create building standards to guide the minimum architectural feature set.
Similarly, accessibility of a Web site is not a property of the bits, nor is it feasible to design them to address every challenge faced by users - instead we have nascent standards for building-in accessibility by proxy, as a set of architectural features.
When we build our town halls and commercial buildings to ADA standards, what we are saying is that we expect certain patterns and relationships to be exhibited in the resulting structure. What we are not saying is that the door frames were field modified to accommodate an Acme brand Model 68020 wheelchair. Nor are we saying that the door frames were field modified because the doors were found to be constructed of defective alloys. The first case is too specific to one assistive device, and second case is clearly cause to demand a completely new set of standards-compliant doors.
Changing style sheets and muddling with markup in order to make something "accessible" for a specific set of browsers is ludicrous. It also smacks of being way too touchy-feely, relying upon subjective qualitative ideation rather than quantitative assessment.
Either we develop to standards, or we don't. Hand waving about pixels being inaccessible, points being "just for print," or percentages and ems being superior, is missing that argument entirely. If some feature of the site doesn't work for you, to the extend it is specific to a given browser it is not an issue of accessibility but a defect in the product... and we will all be better served if we stop making accommodations to defective products.