Supposed to be? Supposed to be by whom? Thickman?
In fact, the exit polls in '04 were fraught with error for reasons that are well understood by political scientists.
The vast majority of poll checkers were college students and they had way too much leeway in picking interviewees. As a result, the exit polled universe in '04 skewed heavily towards the kinds of people who would be eager to talk to college students -- a population considerably more Democratic than the overall vote universe.
There have been numerous papers written on this.
That error is unique to face-to-face exit polling, and has nothing to do with telephone polling. In fact, the telephone polling in 2004 was, on the whole, extremely accurate.