is fastest (about 6× faster)

Actually, you can’t make such a blanket statement. It might be 2 or 3× slower if the dataset is small, like the sample dataset in the OP’s post (I didn’t bother to check). But it can also easily be 100× faster if the dataset is humongous. Results will vary by dataset size, so you can’t give numbers.

Another aspect not to be forgotten is that it always takes twice the memory.

Makeshifts last the longest.