One of the major criticisms of any of the "nonverbal" tests (which by the way aren't quite as nonverbal as you think: the presentation is nonverbal, the response is nonverbal, but the most effective strategies for most kids on matrix reasoning tasks are highly, highly verbal) is just that: If we think of "IQ" (whatever that means) as a sample of a wide range of behaviors, then we would think of a test which only sampled a small range of behaviors as less comprehensive and more likely to identify only those kids who best matched that particular test.
Basically, Raven's and NNAT are not designed to give you the kind of comprehensive information a broad-based IQ test is designed to give. But no pair of tests is going to be perfectly correlated.
