It also needs to be part of any regression testing against new releases. Doing it once against current code does nothing other than say "right now we're okay". I know. I've personally been burned by assumption that what was tested previously is assumed to still be good now.