The first detail to note is that the POVERTY variable depends on family income, rather than household income. So, instead of HHINCOME, you should be looking at FTOTINC. The issue you observe, however, is largely independent of this detail. Therefore, the second note is that IPUMS has introduced some new family interrelationship variables that allow for a more inclusive definition of primary family units than used by the US Census. So, since the values for POVERTY, for the first family unit, are provided by the US Census, this variable uses the US Census defined family unit. By contrast the, FAMUNIT variable uses the IPUMS defined family unit. In the case you observe, the fourth member of the household is related to the household head by being an “unmarried partner”. The US Census classifies this person as being part of a second family unit, while IPUMS classifies this person as being part of the first family unit.
I hope this explanation is helpful. This is a bit complicated and needs to be described better online. These updates should happen soon.