I found this interesting article on the Harvard Business School Working Knowledge site. It examines 13 pay-for-performance programs that various groups in HP tried and dropped. There are lessons to be learned from these experiences.
What gets incented gets done, at the expense of the rest of the stuff. Think about the side effects of any incentive program. Giving extra pay for outstanding individual accomplishments is not going to encourage teamwork, for example.
Manage expectations; better to start with goals that are too high than too low. If management finds that fewer people than expected are able to perform up to the incented level, it is a simple matter to adjust the level down a bit next cycle. If, however, more people than expected make it the first cycle, they are likely to cry ‘Foul!’ when management adjusts the levels up next time.
Creativity and learning are tough to incent with a pay-for-performance program. Such programs tend to reward ‘success’ and results, not experimentation, trial and error.
I agree that these programs can be hard to get right, and are costly when they are done wrong. In management, as in engineering, it is often the hardest things that are most valuable. Pay-for-performance is a tool. It is management’s job to decide if it is the right tool.