I think the idea of the quality rating is fine, it's just that GPU-Z doesn't seem to be able to read/calculate this particularly well.
We talked about this in another thread as some people had over 100% which is a bit ridiculous and others who went from 70% to 80% with the latest release of GPU-Z so I'd take anything to do with the asic with a pinch of salt, at least until this can be found to be a bit more reliable.
the over 100% problem was to do with GPU-Z not being updated for new GPU's but being used on those GPU's... basically all they are doing is reading a value from a register on the card and then converting it to a number - if they change the way the conversion is done based on new information (I wouldn't imagine they are getting much in the way of support from AMD or Nvidia on this) then obviously it will change
GPU-Z is not "testing" anything to come up with the number, purely reading a value and displaying it
having said that - different people have come up with different explanations on what the number ends up meaning for people, but as soon as you say "low means this and high means the other", someone pops up and says "but mine is low and I get X"
there might be a trend (which the makers of GPU-Z have mentioned), but there are always outliers that don't conform to this
the trend seems to be that low asic figures mean higher OC on water, and high ASIC means good OC on air... but it depends on where the faults are on the chip as to wether this is true...
a quality test might come up with a low figure because the chip is really bad in one spot but generally good - this chip will have difficulty OCing at all because it has a weak spot that will give up before the rest of the chip wants to
equally the chip might have a high ASIC % because most of the chip is really good, but has this bad spot - on average the chip is "good" but the weak spot still means it fails early
any test that results in single number to try to cover the entire chip will be averaged and as such an average can be hiding something or skewed in a particular direction by an odd result somewhere in the actual data