My guess is that most fraud is commited in order to get funding, and possibly to get tenure. Sometimes experiments work as you expect, and many times they doesn't. However, in order to have a sucessful scientific career, you must consistently publish papers (with interesting and compelling results) and you must consistently get funded. That schedule doesn't match the reality of good diligent science. As funding levels are at an all time low, it's no wonder that people are getting sloppy and dishonest. I also feel that most fraud isn't 'all out fraud', but results in not repeating experiments and reporting them as repeated, or aggressively dropping data points as 'outliers' with the assumption that further experiments would average them out.
Imagine that you've run an experiment, and need to get a grant before the next year in order to keep your new tenure-track job. Your spouse finally found a good job in this new college town, and your kids are finally adjusting to the move. Oh, and you have a new house and mortgage. Do you not submit the data 'as is' and rerun the experiments, not apply for funding and jeopardizing your job, or do you send the data out because the results look good and you're pretty confident of the findings?
If anyone's interested, the WSJ article the author is skeptical of can be found here: http://www.physics.ohio-state.edu/~wilkins/science-fraud.htm...