I think there's a balance to be struck there. On large databases, upgrade time can be very slow. Is it worth upgrading iteratively at great expense if there are no benefits to be gained (e.g. no bugfixes that affect you, no critical security issues)? Maybe, maybe not.