It strikes me that performance gains associated with sharing address space & stack are similar to many performance gains: trade-offs. So calling them "good" and "evil" when performance is seemingly your sole goal and interest seems a bit forward.
Why do we do this? Well, because it adds emphasis, and a dash of humor.
Clearly fork() is neither Good nor Evil. It's morally neutral. It has no moral value whatsoever. But to say "fork() is evil" is to cause the audience to raise their eyebrows -"what, why would you say fork() is evil?!"- and maybe pay attention.
Yes, there is the risk that the audience might react dismissively because fork() obviously is morally-neutral, so any claim that it is "evil" must be vacuous or hyperbolic. It's a risk I chose to take.
Really, it's a rhetorical device. I think it's pretty standard. I didn't create that device myself -- I've seen it used before and I liked it.
"Good -vs- evil" is obviously hyperbolic - particularly the latter - but outside of morals they still imply a tendency to be technically/practically good or bad in an objective sense. So discounting it as a mere rhetorical device seems overly dismissive.