This probably works as intended, even of it’s not the desired behavior. Given that the protocol only allows clients to activate() the surface that needs attention and never mentions the actual window focus, it is up to the compositor (in this case Kwin) to decide what to do exactly with the request.
On a side note, I think this behavior should be configurable. I for one would not want windows to automatically focus ever (and Plasma even switches the virtual desktops when that happens), but I also see why someone would want to.