Instead of trying to get rid of questions that are gameable, the psychometrists design the test with statistical consistency - things as obscure as, say, what color the candidate answers in a multiple choice question out of orange, yellow, blue and red can be used to correlate with other questions (not literally, but to a candidate it would be equally obscure and seemingly innocuous as a question). If a candidate answers one question in what they perceive to be the societal ideal, this will be exposed in other questions where it's not possible to cross-consistently game the question unless you've read and thoroughly understand a manual of psychology or psychometry.
Tests designed this way allow for a certain amount of "gaming" by candidates before it crosses a statistical threshold, at which point it essentially tells the proctor, "The answers are so inconsistent that the candidate wasn't being honest." and you have to throw out the whole test (which in the context of hiring, means rejecting the candidate).
The reason why this works is because while people will answer "Would you consider yourself hard working?" with "Always" or some other unrealistically gamed superlative, they won't realize that other questions that are seemingly unrelated are highly correlated with that quality - if you answer yes to one and no to another, you probably lied on the transparent one, with high confidence.
Tests like this[1] have been around for so long that they consistently get better, though there are some valid criticisms of them being easier for non-minorities (for a variety of reasons). They are used in clinical and professional contexts, and while the quality personality tests will theoretically be consistent across multiple-test takings (in other words, results are relatively immutable), in practice you should probably avoid giving one individual a lot of exposure to the same exact test twice.
Hope that helps.
[1]: http://en.wikipedia.org/wiki/Minnesota_Multiphasic_Personali...
