Waterfall as a bad idea?

In a past life, the development of hardware and the associated software was done in small incremental cycles. Mary Poppendieck's YouTube talk has some interesting comments about the history of leadership in software development.
    In this past life we built embedded real time systems for missiles and radar systems. While there would be a high level capabilities document along with some interface specifications, the development of the hardware and software tool place in small steps. One primary reasons was we learned very quickly that bugs in Macro-11 and 2901 microcode were wickedly hard to get out if you didn't find them right away.
    When the anti-waterfall proponents speak they usually reference Taylor - which of course a manufacturing management paradigm. So the problem starts when a manufacturing process is applied to an engineering process.
    Watch the video and see what you think. Have we lost the connection with the source of the problem?