Question about iterable dataset and streaming

In the doc, I found the following example: https://github.com/huggingface/datasets/blob/611f5a592359ebac6f858f515c776aa7d99838b2/docs/source/stream.mdx?plain=1#L65-L78

I am confused,
1. If we have already loaded the dataset, why doing `to_iterable_dataset`? Does it go through the dataset faster than map-style dataset?
2. `load_dataset(streaming=True)` is useful for huge dataset, but the speed is slow. How to make it comparable to `to_iterable_dataset` without loading the whole dataset into RAM?