If si_code is equal to CLD_EXITED, then si_status holds 
the least significant eight bits of the exit value of 
the process (with the remaining bits unspecified);
otherwise, it is equal to the signal that caused the 
process to change state.

I can't think of a reason that 32-bit exit statuses would be problematic for the kernel offhand.  There's not internal kernel ABI to worry about in Linux.  RHEL might have a problem, but life is hard for RHEL.

The libc question is... trickier.  I think the realization of this change in glibc would turn on the precise language around si_status. The POSIX manual page (my copy of 'man 3p waitid' anyway) doesn't mention si_status at all.  The Linux manual page says:

 si_status   Either  the  exit status of the child, as given to _exit(2)
             (or exit(3)), or the signal that caused the child to termi‐
             nate,  stop, or continue.  The si_code field can be used to
             determine how to interpret this field.

On that basis I think the text in:

http://austingroupbugs.net/view.php?id=594#c1317

is plausible, since it basically says the same thing, look at si_code first.  There might be binary-compatibility reasons to want to version the libc symbols for waitid() and sigaction() so that old apps get precisely the same bits they always did while new apps are expected to get it right, but that's a glibc maintainer decision.