A poll involves taking a small sample of eligible voters and projecting an outcome based on a likely voter model. A poll can be off for a number of reasons, including (A) failure to represent the demographics of the voting age population, (B) failure of the likely voter model to represent actual voter turnout, and (C) failure to represent intensity of support for a candidate or propensity to change. Overlaying all of these is the consideration of sample size,which is inversely proportional to margin of error. That, more than anything, is why polls vary from actual results and why aggregators like Nate Silver use multiple polls to develop their projections.
The tally of super-delegate allegiance is a completely different proposition. First of all, it’s a census-based count, which means it surveys 100% of the population (i.e. all super-delegates). So, by definition, there is no margin of error derived from sample size. Second, all super-delegates will participate in the nomination process, so there is no reliance on a likely voter model. Third, the historical record shows that super delegates are extremely unlikely to shift allegiance once the state primary calendar has concluded. The best evidence for that is 2008. Supers did shift to Obama once it became clear that he’d have a majority of pledged delegates, but their allegiances didn’t materially change after that. For these reasons, the 10.6% number is completely irrelevant to the super delegate process….but I suspect you already know that.