I'd also suggest that using a copying sweep would be the fastest method still. That is, as you iterate the arraylist, copy the elements into a new arraylist that are still "live", and then switch to that arraylist, like a double-buffer. This should make absolutely the best use of caches.

Cas

I've only ever used this method for particle effects where there are hundreds or thousands of them created and deleted each frame. It's cheaper to pull ones already allocated off a queue than it is to deallocate them and reallocate new ones every frame.

Actually Apple's UIListView does this as well. This is their UI element when you are scrolling through a very long list of elements (like the address book), it only ends up allocating a few and changes their contents as you scroll through them. This however is because you only have a few on screen at once (like 10 maybe) so it doesn't make sense to allocate 200 of them. But you don't face that issue at all, really, and enemies on screen are not nearly as predictable as a list view.