I think what they're arguing is that since most engineers don't use the algorithmic knowledge on a day-to-day basis (and are thus likely to be rusty), the time they'd need to spend reviewing/prepping to pass a whiteboard coding interview is greater than the time it would take to do the work sample.