It's funny. When you first do work, you want the experiment to satisfy your hypothesis. When you are building on work, you also want the replication to succeed. But when it is a famous result like this, you actually want it to fail so people talk about your result. There are uncountable ways that these experiments can be unconsciously and subtly affected by the desire of the experimenter.

As an aside, I believe one interesting confounder in the marshmallow test is that it tests more (or at least as much) the subject's trust that the eventual reward will actually be given as it does the subject's ability to wait for the reward. So if you live in an unpredictable environment, it's better to just eat it.