Any production where you're routinely scanning a text file with a million records is probably a batch process, and I'd be shocked if the usual performance wasn'

Any production where you're routinely scanning a text file with a million records is probably a batch process, and I'd be shocked if the usual performance wasn't much closer to the worst case than the average.