As you have discovered, either record can be way off. The reason lies in the fact that both records are often created by a person with indirect knowledge of the actual birth.
In the case of the Death Record, the information often is filled in by a child or spouse of the child. Often the information provided is by even more distant parties who have limited or no information on the actual birth event.
In the case of the Census Record, the information can be filled out by the head of household or a neighbor or a child who happened to answer the door. Then it is further obfuscated by training of the Cenus Taker and their attention to detail. The Census is nortoious for wide swings in birth years.
In both cases the information for a Birth is a second hand source. Being provided by persons who have no direct knowledge of the event. Unlike the Death date on the Death Record where they may have actually witnessed the death. That being a first hand source and thus more reliable.
So neither is good or better but you might need to dig a little deeper and see who filled in the Death Record or determine who the Census taker obtained the information from. The 1940 Census actually records who provided the information for example. It can vary by year and location.
Thus obtaining additional sources can help you lean toward a likely date.
But if you were to force me to choose, my experience would be to go with the Death Record. But that is only slightly better than a guess. You are better off looking for additional sources to corroborate any theories.
And of course a first hand source is the holy grail.