Alex points to an interesting paper, Exploiting moral wiggle room: Experiments demonstrating an illusory preference for fairness :

Subjects in economic experiments are often generous. This behavior is often interpreted as reflecting a preference for equitable, efficient, or otherwise desirable social outcomes. We show that a considerable proportion of such fair behavior may be driven by a desire to appear fair without actually wanting a fair outcome. To do so, we first demonstrate a high frequency of fair behavior in a modification of the standard dictator game, but then show that fairness decreases substantially when the connection between choices and outcomes is obfuscated.

Specifically, we show that in a binary version of the dictator game, a majority of subjects choose the fair and efficient outcome. We then show that subjects playing the same game instead choose to maximize their own payoffs, at the expense of fairness and efficiency, when the recipients’ payoffs are uncertain, even if this uncertainty can be costlessly resolved. We also find that when either of two subjects can sacrifice to implement a fair and efficient outcome, but neither can ensure fairness or inefficiency, selfishness prevails. Finally, we also find less fair behavior when unfair outcomes can plausibly result from an external factor rather than just the dictator’s choice.

Our results indicate that much fair behavior may reflect motivations other than simply a preference for desirable social outcomes. Instead, much of the behavior in our experiments is consistent with a desire to pursue self-interest, while maintaining the illusion of behaving fairly.

This rings true for me, it's consistent with anecdotal evidence. I've read a lot of those studies that are "often interpreted as reflecting a preference for equitable, efficient, or otherwise desirable social outcomes" and wondered about experimental design as well as selection of subjects - the old starving-students-skew concern. It appears from this study that those earlier ones showed that people are fair mainly, or perhaps only, when it is in their interests, when they can unambiguously be blamed for unfairness. Give them reasonable doubt, obfuscate the connection between choices and outcomes, and their behavior changes.