It seems to me that it would be very rare for an elisp package to be broken on one arch but work on the others. However, as far as I can tell, all the app-emacs ebuilds use the KEYWORDS variable in the regular fashion. This creates unnecessary duplication of work as a package must be tested on each architecture before it can be flagged stable. I propose that all the builds which use the elisp class should replace the KEYWORDS variable with ELISP_STABLE, which could be true or false. According to this value, elisp.eclass would then set KEYWORDS to be stable on all archs, or unstable on all archs. In the case where this scheme is inappropriate, the ebuild could just set KEYWORDS itself, and leave ELISP_STABLE unset. Does this seem like a good idea to anyone apart from me?
While it is true that that many or most of the emacs lisp packages will work on all platforms on which emacs does, some packages may depend on external programs, or have some other issue, and so it has been decided that packages should not be marked even in ~arch if they have not been tested even once on that platform. If you have tested an elisp package on some non-x86 platform, please tell us and we will mark it at least ~arch.
Thanks Jeremy, No reason to leave this bug open.