The comment by nfdc23 answers the question:

"There's no good reason, and in particular nothing pathological for the non-smooth case. Perhaps some paper working with smooth varieties stated the result in the relevant context and someone getting it from there copied the SGA6 reference without reading it and carried over the smoothness for safety, and then it spread due to people continuing to copy the citation from papers without reading the original reference. Near the end of 8.4 in the book Neron Models the result is stated in the same generality as in SGA6 (in particular, no parasitic smoothness hypotheses)."