Just to state the obvious, in practice over the long term, you've got to be deallocating faster than you're allocating, or something nasty that will at the very least violate your performance constraints will happen.
Right. And my point is that the deallocation vs. allocation ratio is the only metric you really need. How fast garbage is made is completely irrelevant because over the long term it's bounded by the allocation rate. You don't have to solve the hard problem of figuring out how fast garbage can be made, you can solve the much easier problem of bounding allocation. And of course in either scenario you have to show that there are no leaks.