Hmm, I don't know the specific details of the test in question but adjusting your behaviour to look good on static frames is a fairly common strategy on video.
If say the video in question is interlaced, then you have several ways in which you can display it.
You can
a) Display the two fields woven. (Known as weave)
b) Display the odd / even fields seperatly, scaling them up to fill the output. (Known as Bob)
c) Perform some kind of adaptive system where you interpolate between weave/bob depending on the characteristics of the video.
Now if you have a static scene then weave is the way to display this as it will provide twice the vertical resolution of a bobbed image and will not wobble up and down the way that bobbing can do. If however the scene is moving then weave will produce combing artifacts that look quite unpleasant, in which case bobbing is the way to go. If you have the processing power then you can determine which to do on a per pixel basis which will provide better results than either.
There are other methods out there as well, but this demonstrates that adjusting behaviour if the image is static is a perfectly legitimate thing to do. If the result of this causes other artifacts then it is the fault of the test that it does not pick these up, not the fault of the gfx card/drivers.
CC
