Nope, I've turned off online metadata. I've debugged it and the issue is that Kodi wouldn't pipeline the SMB read requests, so a huge chunk of time is spent waiting for small bits of data to arrive.
My last info was, that we don't process anything in parallel, as the old xml scrapers were not optimized for that. As we're moving to python scrapers now, we should be able to optimize the interface to handle more in parallel. That was the plan at least, but XML scrapers first need to be removed, as they hold back the API.