The reason why this poll is different is that they allowed respondents the choice of English or Spanish. Nate Silver points out that the two states which he called wrong in 2010 were NV and CO, both with big Hispanic populations. So it is possible that all polling is underestimating the Hispanic vote.
http://fivethirtyeight.blogs.nytimes.com/2012/10/13/oct-13-arizona-and-the-spanish-speaking-vote/He then ran a version of his model as a thought experiment on what if Obama actually did win the Hispanic vote by 70..
While it doesnt actually put AZ into play it was a big enough difference to flip VA, CO and FL from lean Romney to lean Obama.
Im not sure that is going to happen, but as I've noted before, the biggest factor in polls is not party ID weight but white/non-white weight. This is just more of that theme.
This issue of polling in Spanish may be something that pollsters need to deal with just like with cell phones and the youngs. Two big Democrat demographics are hard to reach with traditional landline English polling.
