https://github.com/protocolbuffers/protobuf/commit/50f9ac3ba433e7be5869ee8ec7e7a85e6ca09a3d https://github.com/protocolbuffers/protobuf/commit/50f9ac3ba... On

https://github.com/protocolbuffers/protobuf/commit/50f9ac3ba433e7be5869ee8ec7e7a85e6ca09a3d https://github.com/protocolbuffers/protobuf/commit/50f9ac3ba... On a sufficiently anemic CPU doing enough fast allocation you can get measurable improvements from that kind of thing, but for a superscalar CPU and a general purpose allocator I agree you'd be hard pressed.