I would like to add to “cons” that there is no built-in support for spatial indexing, making exact raycasting, frustum culling and collision detection hopelessly inefficient in complex scenes.
I even think the complexity/maintenance argument against adding such a feature would be put to shame by all the power users who would gladly spend time ensuring it works for their project.
See e.g. Fast raycasting and other spatial queries and https://github.com/mrdoob/three.js/issues/13909