I'm not sure how that's relevant, misunderstandings of your post's parent aside.

On the subject of unsupported browser tests, a question: how much value do you actually derive in real-world browsing from SMIL, the remaining area of Acid3 not supported in Firefox? (Aside from being able to run Acid3 and see it pass, that is.) How much do you expect you would derive if SMIL support were widespread in browsers? I think the answer is very little, and I think my opinion deserves at least some consideration given that Ian has repeatedly stated that adding the SVG-related tests to Acid3 was a mistake. (I don't know that my questions touch upon the reasons Ian has for thinking that a mistake, but I do suspect unusefulness in general, even in an ideal world of SMIL-supporting browsers, played a factor.)

Acid2 was good in my book, but Acid3 was much more of a grab bag. It mostly consisted of browser bugs found by searching through bug databases and through suggestions to Ian (I suggested a few, actually, based on knowledge of various Mozilla bugs), not something that deliberately set out to test functionality useful to web developers in general. Some of the fixes were utterly trivial (I fixed one that was a one-line change); others were more involved but not particularly difficult (I fixed several of these as well). Really, tho, the fundamental problem was that Acid2 tested things developers wanted to use but couldn't; Acid3 tested many things which will never see much use by developers.

Actually, passing Acid3 at this point apparently means supporting the standard wrongly

Not really; if you support CSS level 2, 2.1, or don't support the CSS3 backgrounds and borders spec, the reference rendering of Acid2 is fully correct. The problem is Acid2 made an assumption that a standard CSS2.1 property would never be extended in a particular way by future CSS levels/specifications, but it was. If you support CSS3's current definition of background-color, then Firefox's rendering is correct. If you don't, then the reference rendering is correct. So really, latest Firefox and Safari are both correct.

The problem isn't that anyone is or isn't failing, it's that Acid2's pass condition is not correct independent of implementation of future CSS specifications. It's not like this could really have been predicted, so it's more an odd quirk of the standards process than anything else.

Incidentally, the spec might still change on this, partly because Acid2 would be broken and partly because the fallback color in CSS3 that causes the problem here is a bit of an oddball feature, both conceptually and syntactically. If it stays the fallback color might move to, say, background-image or require a disambiguating slash or something else to avoid this edge case (and not "regress" Acid2 to show red in compliant implementations).

I think that illustrates why we shouldn't rely on tests like the Acid Tests too much when determining standards compatibility.

I'll agree with this. The Acid tests demonstrate support for precisely what they contain; they imply support for much more than that, since nobody really thinks a full stack of Acid-specific hacks wouldn't go unnoticed if it were attempted, but unless you wrote or reviewed the patch it's going to be hard to find the unaddressed edge cases that the changes might not address in the relevant functionality.

Nothing really demonstrates standards compatibility. However, a lot of tests of simple uses and complex uses in which multiple features interact in complex ways go a long way to showing that the particular features probably work correctly in practice.

(Oh, I wrote the post you linked, if my username didn't make that clear. Good ol' referrer tracking for the win...)