[bunderbunder]

The ones described in this article are ones that shouldn't necessarily be constrained to tight loops if you're looking for maximum benefit. Because their real gain isn't saving CPU cycles, it's reducing load on the memory manager. There's definitely still a diminishing returns situation, but the returns may not diminish as quickly since a memory allocation done by one spot in the code has the ability to impact performance virtually anywhere else when you're working in a garbage-collected language.