Comment by jbapple

When I benchmarked that a few years ago, I found the overhead of a syscall made mmap much slower (assuming the C or C++ memory allocator had already received some memory from the OS it could dole out).