while trying to re merge some of our company local ebuilds, i realized that having set -e in any phase makes portage exit with an error after that phase completed (successfully!). this wasn't the case some revisions back (ok, ok, it's been a few months, since i last tried ;))
my wild guess is, that there was some fork-optimization and a subshell has been removed, which would have reset the set -e (or better, the set -e would not have affected the outer shell - i.e. ebuild.sh).
i'm removing the set -e's from out ebuilds, but still it feels like a bug that setting shell options influences ebuild.sh...
I don't know why you've noticed a behavior change, but in svn r13273 I've made it call set +e in qa_source(), qa_call(), and die(). A solution involving a subshell would be significantly less practical since we'd have to save the ebuild environment from inside that subshell, and we don't want that code to have to assume that set -e might be enabled.
This is fixed in 2.2_rc29.
This is fixed in 184.108.40.206.