When you introduce a HF layer in your environment, you offload all the ingestion-phase work, except the actual indexing onto HF's. So your indexers don't have to bother with listening for HEC events, doing time extraction, indexed field extractions, ingest-time evals.This leaves your indexers for the sole purpose of indexing and searching.
Additionally, there are apps which do not run on UF's but need indexer or HF. These are typically apps running some scripted/modular inputs. Running them on indexers would add unnecessary asymmetrical load if you ran them on a single indexer in a cluster.
So as your environment grows introducing a HF layer has its pros.
EDIT: Oh, and one more thing. Two actually.
If you have an intermediate load-ballancing forwarder layer you don't have to update your outputs on UF's when you grow "sideways" adding new indexers. It's annoying especially if you don't manage UF's with deployment server. You're more likely to need another indexer or two than new HF's.
And if if you offload most (if not all) ingest-time apps to HF, you don't have to restart your indexers if you have to change something in them.
