There are a number of threads in the forums that provide examples on how to get the iteration count. This is just one
example. You could use something like that to get the iteration number and bind it to a parameter. Then use that parameter as part of a conditional execution on a test case/smart folder. So in effect, execute when iteration count > 2 or something.
All in all this doesn't seem like the most clean and clear way to design a test case; if/when you dataset changes you're going to have to adjust this condition. However you may have situations that make this the best possible solution.