Comment 18 for bug 36538

Given that you can't actually get a 'slow' amd64 machine and that whatever performance increase -O3 gives is probably negligible (and comes at the cost of stability, clearly), this doesn't seem like reasonable argument against this, for this architecture?

I've been rebuilding the packages since discovering this and have noticed no appreciable drop in performance (but a very appreciable improvement in the whole not-crashing department)