This is totally wrong.

Read the comment you linked more carefully. I'm not talking about reality -- I'm talking about the model which RichardKennaway proposed. Specifically, I find this model too simple because it has a single force acting on the spread of a gene -- the "selective disadvantage D". Note, by the way, that it's not about neutral mutations at all, presumably D is not zero and we are talking about mutations which are actually selected against.

Given that, the expected value of X (I'll grant you that I should have been more clear that I'm talking about the expected value and not about what one instantiation of a random process could possibly be) must decrease.

There is no question to beg

In the model proposed there is.