It's common knowledge in sabermetric circles that it's not just the stolen base that's important, it's also the caught stealing. A base stealer is only valuable if the impact of his successful steals outweighs the times he's caught.
The rule of thumb is that you need to be successful close to 75% of the time in order for an attempted steal to be worth it. This is determined by looking at the expected run value of the success and the failure and calculating what is known as the breakeven point.
Expected run value is an okay way to figure out the breakeven point, but a better one is by using win value. Using win value takes into account the fact that stolen bases attempts generally occur in situations that are more important than average.
Update: I've had at least one request for the raw data, so I've posted it here for all that are interested.
As part of the preparation for my historical stolen base article at The Hardball Times, I calculated the actual average win value of both the stolen base and the caught stealing for each season the Retrosheet era (1954-2008).
This allowed me to figure out the breakeven point for each of those seasons and plot them on this graph.
Be a little bit careful in quoting these numbers though. For this to be the true breakeven point, we'd need to assume that managers and players have figured out the optimum times to run, which could be a poor assumption.
The proper way to do this (I think) is to figure out the breakeven point for each possible steal situation and then calculate the average by weighting how often each situation occurs.
My guess is that the game on the field is closer to efficient than many like to think, so I think the actual numbers will be fairly close to the right answer.
References
The Win Expectancy and Leverage Index data is licenced from http://www.InsideTheBook.com
The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org".