In the absence of  a "preference" factor, the tendency is  to  keep  the  older  NAR.
Nonetheless, in the case of  the double NARs  for Gavri'el Tsoren (n88-241404) and  Gabriel Zoran (nr91-12990),
LC agrees that, since all other factors are equal (there's one  item in LC database under each heading, for example),
the record with the non-systematic romanization is  preferable.  we're  therefore deleting the 1st NAR and updating the 2nd.