Efficient markets advocates often highlight the asymmetry in publication results: if you find an empirically improbable result, where improbable is defined relative to a standard rational model, you get published; if you find a probable result, you usually do not. Since the set of all possible violations of standard rational models is large, you should expect to see many "violations" of efficient markets even if there are not any. This is because improbable is a statistical concept, and if you do 1000 tests against a true null hypothesis, we should expect 50 of them to be significant violations at the 5% level. Lo and MacKinley (see Lo's pubs here) and White (see his publication links here), have tried to explicitly correct for this data snooping bias, but in my opinion the net result is still very qualitative: be skeptical of non-standard findings. This is not to say that all violations are spurious, only we need an appropriate amount of skepticism when we see any new result.

This is not restricted to finance, of course. Medical research has the same problem. CNN notes (see here) that many findings are subsequently debunked:

[A] review of major studies published in three influential medical journals between 1990 and 2003, including 45 highly publicized studies that initially claimed a drug or other treatment worked.

Subsequent research contradicted results of seven studies -- 16 percent -- and reported weaker results for seven others, an additional 16 percent.
...
Editors at the New England Journal of Medicine added in a statement: "A single study is not the final word, and that is an important message."

The refuted studies dealt with a wide range of drugs and treatments. Hormone pills were once thought to protect menopausal women from heart disease but later were shown to do the opposite, and Vitamin E pills have not been shown to prevent heart attacks, contrary to initial results.