You're going to have a hard time finding numbers, and even if you do, you'll need to define your terms better before you can understand the numbers. One man's cowboy is another man's agile developer.
Also, "fail" is ambiguous. To me, a failed project is one that never makes it to market in the first place, so by definition there are zero failed projects in the marketplace.
Finally, cowboy coding is never the reason that a project fail, in the sense of being cancelled or abandoned. It's one factor that can lead to the reasons that projects are cancelled, along with poor management, lousy process, lack of skill, lack of funds, legal issues, and so on. Nobody ever says: "I'm canceling the project because Sixgun Slim over here can't be bothered to use our development process," but they do say: "I'm canceling the project because the schedule has slipped four times already, we're creating bugs faster than we're fixing them, and my judgement is that it'll be cheaper to stop now than to stop three months from now."
So, if you want to know how often cowboy coding leads to project failure, you'll need to first look at the real reasons that projects fail, and then try to determine how often those problems are caused by individuals who ignore process. Don't forget to also look for cases where projects that are failing due to over-restrictive processes are saved by a visionary maverick who gets things done by flouting the process. It's been known to happen.
