Well, the link posted in #4 goes to an article that quotes Zogby, so that explains that one! I don't believe a word Zogby says anymore.
Well, one reason is that the campaigns are willing to pay the extra dough for a higher sampling rate - say, 1500 voters instead of the 800 or 900 a media outlet's public poll does. You get diminishing returns the more you increase the sample size - 2000 respondents would only add a few tenths of a point more accuracy than 1500, but cost 25% more - but up to around 1500 or so it's worth it for the added accuracy.
Also, internal polls filter out the BS. They stick with LIKELY voters, for one thing, not just registered voters, half of whom won't even bother to show up on election day. They also filter out the Democratic bias. And by that I don't mean an intentional attempt on the pollsters' parts to skew things for the RATS, but rather a real, honest phenomenon that causes all polls to come out showing a few percentage points more support for the RAT candidate than actually exists. (The reasons for this have never been proven, though they're believed to be largely due to political correctness. There are still lots of people afraid to say out loud that they support a Republican, lest they get tarred as a racist, homophobe, [insert standard-issue rat mudslinging term here]. So they tell the pollster they support the RAT, and then vote GOP in the privacy of the polling booth.) The public pollsters will not correct for this phenomenon statistically, because that would be unfaiiiiiirrrrr to the Democrats, since they can't really "prove" the phenomenon exists at all. The internal campaign pollsters don't need to play those games. They eliminate the BS and find out what's REALLY going on. And that's why they're more accurate.
And that's also why anytime you see both candidates "neck-and-neck," you can rest assured it means the GOP candidate has at least a 1-2% advantage.