I do reviews strictly on a first come first serve basis, and I will not prioritise anything based on deadlines. As a matter of principle I will not accept anything that gives me three weeks or less time. Not only will this time frame not normally work, also a journal that sets too short deadlines signals in my view that it prioritises speed over quality, and I don't support this.
Other than that, I try to have a maximum of four or five review tasks open at any time, i.e., if I reach this maximum, I start declining. If I am for some reason more busy than usual, I may lower this maximum. Somebody who likes reviewing less than me may generally set themselves a lower maximum of course.
In most periods I get more requests than I can handle under these rules, but if I decline things that are too far from my expertise or that seem of hardly any interest to me, I can usually get things down to a manageable number. (Obviously these criteria can be handled with some flexibility.)
Experience shows that in this way I can do almost all reviews in somewhere between three and seven weeks, and although seven is too long for most dealines, it rarely gets that high, and most editors (who didn't set ultra tight deadlines) are still just about OK with seven weeks. (I am an editor myself and I'd be very happy if I could get all reviews within eight weeks, say, although of course getting many by the actual deadline helps.)