In theory, you should be able to merge the NBER March supplement and the IPUMS CPS supplement for every year. Moreover, I can merge them perfectly from 1988 to 2016 for all years except for 2001 and 2014.
I think it has something to do with special sampling in those two years in particular. In 2001, the CPS also had SCHIP supplement questions and in 2014 the CPS did an odd sampling in order to incorporate new income questions. Can someone confirm that these are the reasons the number of observations differ? If not, can someone explain why this only occures in these two years?