1. It prevents my "flow" from being sidetracked by micro-optimizations that are probably too early to consider necessary anyway.
2. It helps me to retain my short term memory on the code I am working on. If I branched out at each TODO to implement some improvement or method, my brain typically needs to context switch to focus on the fine details of the subject. By the time this is done, and I switch back, I have forgotten key aspects of what I was working on and it slows me down again (i.e. local names, structs, etc.)
3. It allows me to re-approach something with a completely different mind set (given that I come back to it after a signicant amount of times). Half the time I realise that what I wrote was indeed "good enough" and no further time should be committed to it unless a reason exists to do so.
4. It gives opportunity for other developers to see, think, comment and contribute on the subject. I find that typically if I TODO an area, it's good for a second set of eye balls to see it. There's far smarter people than me around, and there's a good chance one of them will find it and propose a better solution.
5. On the rare "quiet" day, I can grep for TODO and just work through them.
Obviously these points are only valid if the TODO labels are being added in situations that will benefit from the above.