Yes but at that point you can just use a memory pool and those are not hard to implement (see the implementation linked in my previous message). The memory pool can be used side-by-side with ARC/ORC and provide speed up for only the type of objects that need them due to heavy allocation/deallocation.
Also benchmarks are kinda moot when we can't see the sources (you mentioned having modified them for regions) as well as means to reproduce your numbers. For the simple benchmark like the one you made, I'd recommend using tools like @disruptek [golden](https://github.com/disruptek/golden) or [hyperfine](https://github.com/sharkdp/hyperfine) to minimize fluctuation in the results due to external factors. And for a memory management benchmark it's important to have the memory usage logged so that we can observe how memory efficient the strategy is.
