I have seen this happen before (but not that often). A possible workaround: use Text+ instead, which involves the Fusion engine, and see if it's any better.
The alternative would be to render out the titles or prep them with another program like Illustrator or Photoshop and bring them back into Resolve as TIFFs or PNGs with an Alpha Channel. We generally request titles as TIFFs for the simple reason that we know that precise line spacing and dropshadows and so on are not ever going to change, no matter what. And we're religious about comparing them to the client's reference video to make sure the titles are absolutely perfect in terms of placement, fades, and so on.