If a state has 5 counties, but COUNTYFIP identifies households in only 3 of these counties, can I assume that the rest of of the observations, with the value 000, are for the other 2 counties combined? Or is there some reason this would not be the case, that some of these values would mean something else?
Your interpretation is correct. COUNTYFIP=0 when a household is located in a county that cannot be identified by IPUMS geographers. Note that counties are not identified directly in the public use microdata from 1950-onward. IPUMS geographers are sometimes able to infer county from other publicly available information, like PUMA. Data users can make inferences about the set of counties that individuals and households reside in based on which counties are identified and which counties remain unidentifiable, within a state.