I think their main complaint is that you can't iteratively operate on a stream as a whole without first converting it to an array, which besides sometimes requiring awkward syntax, can require a lot of memory for large datasets.