It just needs to be read. The reader processes more-or-less push it onto the network verbatim. The writer is not network-aware.
Thinking about this during those epiphanal moments you get while riding the bus, PAGE_WRITECOPY is like a one-shot trigger, so once the copy of the page is made by each reader process, I would need the writer to tear down the shared section and create another in order for it to trigger again, and notify the reader processes of the new handle... :/
I also spent some time digging into the modern Linux virtual memory manager, and when it does copy-on-write (in response to a write fault on what is otherwise a read-only page), it does a memcpy() specifically using SSE instructions on Intel. I'd expect clang (my compiler) to also think about doing this. The main difference would be that it would be done in kernel mode, so might be a teeny bit faster.
