x86_64 malloc already uses 16 byte alignment and portions to support SSE commands. So why not go all the way and allocate full cache lines? This also prevents accidental false sharing between different malloc pointers.