Unfortunately, these HTML/CSS/Javascript tests tend not to be about correctness on the large. ACID3 was probably the best example of something that was meant to specifically show how IE was poor. It tested specifically those items that IE was deficient at. It didn't weigh all aspects of the standard equally.
An ideal test would go through and enumerate the standard and attempt to weigh each feature with expected usage and potential to workaround. I've yet to see anyone even attempt it.
The W3C HTML Working Group is working on a complete testsuite for HTML5. The WebApps working group are doing similar things for their specs. many of which are often loosely referred to as "HTML5". The goal of these testsuites is to help browsers achieve interoperability, so the emphasis is on getting high quality tests that cover the whole spec and include difficult cases.
If you want to help out, contributions are very much appreciated; see [1]. Indeed, contributing tests is probably the most effective way to reduce browser interoperability problems and hence pain for future web developers.
If you want to take the tests and start adding subjective weightings based on importance and ease of working around fails, the liberal (BSD-style) license will let you do that (although obviously you can't claim that your derived work is the official W3C testsuite anymore).
It looks like they're not very far along. Do you know if there is an ETA on when they think it will reach signifigant?
I find it surprising to see they only have 925 tests. It seems to me that a spec of this nature is the type that would really benefit from a very test-based approach (as a s section is being written, so are conformance tests).
An ideal test would go through and enumerate the standard and attempt to weigh each feature with expected usage and potential to workaround. I've yet to see anyone even attempt it.