The attached test program can be used to measure the performance of
posix_spawn vs fork+exec under simulated memory conditions in the
parent. The options it takes are:

-n bytes	memory to allocate
-r		fragment memory into alternating ro/rw vmas
-d		dirty allocated memory
-f		use fork+exec (instead of default posix_spawn)

At -n 200000000 -r -d, I've observed a nearly 300x performance
difference.

As (somewhat) expected, with large -n but neither -r or -d, fork+exec
performs well -- all pages are COW zero pages and easily forked. But
if the pages are dirty, or if there are lots of vmas (even non-dirty
ones), fork gets very slow.

Rich