I completely agree with you that the media’s interpretation of Bloomberg’s recent presidential poll takes its result as being far too close to fact and not enough as an unusual result that is far outside of the expectations set by other recent polls. It is definitely an outlier in that context.
You state that the polls we see are “very, very wrong” about one out of 20 times, which makes some sense when you consider that their margin of error is typically based on a 95% confidence interval. I don’t know that I would describe a 5% chance of a random result falling outside of a range as “very, very wrong”, but that’s a different point. What is still important to consider here is that once a result gets outside of its confidence interval, it is still subject to its distribution. Not all “very, very wrong” results are created equal.
In fact, if we take the Real Clear Politics average of 0.6% favoring Obama as the mean in a normal distribution with a similar sample size (about 1000, meaning a margin of error of 3.1), we find that the likelihood of getting a random result 12.4 points in either direction is less than 1 in 100 Trillion. Effectively, it’s zero. Which means that this poll says something about the national opinion for president. It calls RCP’s average significantly into question and cannot be written off as a simple outlier. It's a pretty big deal.
Put another (shorter) way:
Bloomberg: Obama by 13%
Chances that the REAL national opinion is either lower than 9.9% or higher than 16.1%: 1 in 20
Chances that the REAL national opinion is either lower than 0.6% or higher than 25.4%: <1 in 100 Trillion
It would take an awful lot to get Bloomberg’s result if RCP’s average is really correct or even close to correct. Something is amiss.
Here's more from pollster Mark Blumenthal, who also concludes, "the most likely possibility is that this poll simply represents a statistical outlier." Comments are, as always, open.