Most ECS implementations realistically are high-performance in-memory columnar databases, where systems use granular mutation patterns. Hell, the main access pattern is literally a called a Query. It's become quite a meme to call bevy_ecs "turning games into SQL databases". If you've used BigQuery or Redshift, you've basically just used a super-scaled up ECS engine.

I think to some extent that's still an open question that hasn't been fully explored. I've seen a lot of people suggesting it could work well for application UI development, but nobody has proved it out with a solid implementation yet.