The sample size of Retina Macbook Pros is large enough where one can ignore issues such as replacement. Yes such generalizations can and should be made when creating simple statistical models.
Regardless that's not what it looked liked your post implied (and I'm not saying that is explicitly what you wanted to convey). It looked as though you were stating that because he got two faulty machines there was a significantly lower chance of getting a faulty third one.
Honestly though, the chance of getting a faulty machine are pretty low and the OP is more unlucky than anything else.