Mason is right that if Obama only got 76% of the black vote while Romney got 20%, that would be a record breaking bad result for a Democratic candidate among that demographic. Also, if turnout among black people is the same in 2012 as in 2008, this difference in black support would mean a loss of about 2.86 million votes for Obama, while it would mean a gain of about 2.54 million votes for Romney. In other words, a net loss of 5.4 million votes for Obama. All else being the same as in 2008, this alone wouldn't have been enough for McCain to win a majority in the popular vote, but all the signs are pointing towards 2012 being a significantly closer election than 2008. In short, an election in which the black vote split 76-20 between the candidates would be a disaster for the Democrats, especially in this election.
BUT, as mentioned, these poor polling numbers seem almost too bad to be true, so I would be very reluctant to take them at face value until more polling firms find similar results. This might just be a statistical outlier. I also don't find the conclusion that black folks are responding to Obama's support of gay marriage all that convincing; yes, black folks are statistically less likely to support gay marriage than Americans on average, but not by enough to explain a sudden 19 point drop in support. Furthermore, I don't think that anything short of Obama coming out in support of slavery would cause support to shift in the Republican direction the way this poll predicts. What's more likely, I think, is that if black people are dissatisfied with Obama, they will remember on election day that they also aren't that particularly fond of Republicans either, so they'll probably just stay home rather than vote for Romney.
If the poll had predicted a return to the historical norm of 80-90% support of the Democrat then I might have been able to buy it. As it stands, it just defies reason that black voters would abandon the first black President in droves like this.