I've read this article on the topic and here is the conclusion:
"Anyway, what I’ve tried to demonstrate was usage scenarios that you’ll need to consider for your own real cases: INT remain smaller in storage (50%) and will only perform better if INSERTs and SELECTs are already fed with an INT value - and this is specially relevant for WRITE-intensive scenarios - but DATETIME alleviates extra responsability/care from the developer. Programmers don’t usually care about this, and want the most flexibility from the database, so it’s up to you to find with them a compromise. I may have provided both enough arguments for an endless discussion, though"
However, what kind of DB are we talking about in your case? Do you have the thousands of writes and reads and millions of records, that justify switching from DATETIME to INT? Are you sure you don't produce a lot of overhead when you start date comparisons using ints?