That's why I mentioned "doesn't make financial sense" too. If you can throw another $X/mth box at it, or have engineers spend weeks improving the system's performance for the same effect... there's a point where scaling out makes way more sense. Whether that leads to complexity issues is really case-by-case.