The problem is that even the most test-loving psychometricians know the limits of testing. They understand that even the best tests are merely samples of a larger domain and making inferences based on these tests must be tempered with the understanding that all tests have an element of error that must be compiled with multiple measures.
The problem with census testing is that we get cocky and reckless with the results. Just because everyone writes the test that doesn't change the fact that the test is still nothing more than a sample. Those who are not familiar with testing seem to think otherwise.
While it is true that I am not a proponent of testing, even I can see how moving from census to sample testing could be a step in the right direction. Here's what I mean:
When we hear the results of political polling, we would never say "well, if we have polling, why do we even need to bother with the election?" When it comes to polling, I think even the most uninformed understand that polls are merely samples of the entire population - polls may be informative but they are not a substitute for the actual election.
If testing was done only by a sample of the population, I think it would be much harder for anyone to sell testing as a substitute for other more authentic and direct measures of learning. Sample testing would help us all to understand that the tool chosen to tell us about our schools cannot do what we ask of it.