Why is posix_spawn faster when its just a wrapper around fork + exec?

Question

This is actually a 2 part question.

1. Why is posix_spawn() faster than hand written fork() + exec() when posix_spawn() is itself just a wrapper around fork() + exec()?
2. Why is posix_spawn() with POSIX_SPAWN_USEVFORK slower than normal hand written vfork() + exec()?

According to Ruby posix spawning library work it seems like posix_spawn() is giving consistent throughput. I read the source code of glibc, but I couldn't find anything obvious that is giving this consistency. Can someone explain it to me?

I made a small test source file (too big for here, so GitHub), and had it run about 30 times - each time invoking /bin/true several times (random number between 1000-10000) - then dividing the total time consumed by total iterations. The results are shown below. Its run on an average Linux Mint desktop with no bells and whistles attached.

I gather the man page considers vfork() to be an architectural blemish, but for my use cases, its just fine.

Here, the v prefix is using vfork() and n prefix is the normal fork().