Rebuild on a usable index is performed by reading the old index (unless compute statistics is invoked, i believe). Recreating it is performed by reading the table, of course ... so would you rather read an index twice, or read the table once?