The "overclass" (continuing with your phrasing) is also dependent on various forms of social welfare. Most recently through socialized business losses and corporate bailouts (qv: TBTF, zombie banks, federal discount window rates), on complex tax policies (qv: "double irish", etc), on expenditures on "nation building" and "police actions" and related distortions, and indirectly through qualitative easing and depressed exchange rates and policies that can move jobs and capital offshore.
And that's before you get to discussions around the mortgage deductions and other forms of social welfare, which are aimed at homeowners and particularly at the rates that banks can charge.
Or at whether or not there will be any number of prosecutions for securities fraud, or investors in failed entities getting their hair cut, or around that national candidates can measure campaign contributions in the multi-millions or now billions of dollars, or any increase in market transparency or the effects of HFT on the market. (This is probably the most pernicious form of social welfare, as a crooked system leaches out the money; business continues, but a lack of trust costs everybody.)
And while I doubt that these "overclass" social programs will see particular cuts, this largesse can also lead to social unrest. (Though usually not among the recipients of this largesse.)
Is there a simple answer? No. But a discussion of "underclass" and "social welfare programs" without a corresponding discussion of "overclass" and "corporate welfare programs" and related is playing directly into the divisive political kabuki, and not likely working toward a solution.
I don't see any historical evidence for the claim that stability typically accompanies lack of social programs.
Perhaps it is possible to perform a statistical analysis that shows that the absence of social safety nets does still, in general, produce stability, but I haven't seen one put forth. The linked article certainly does not include any data relevant to that hypothesis.
If that is the case, then the number of riots is related more strongly to withdrawal of benefits than to their base levels.
This would support joshuaheard's claims, though of course it doesn't rule out the possibility that riots(social benefit levels) is a nonlinear function, increasing near 0 and flat in the regime where the original dataset lived.