that's for you to decide how many percent of reserved instances out of total compute power you need. it's always a trade off. in that case, pct of cost saving is lower but availability is higher.