Amen to this -- at Fermilab we had a setup with lots of 8mm tapes, and we thought the MTBF was awfully high (several were failing each week), much more than the 30,000 hour MTBF specified...until we realized it was 30,000 hours with a 5% duty cycle, or 600 hours of use. 600 hours divided by even a dozen tapes is 50 hours, about 6 8-hour days and these were in use up to 24x6... The system also let us empirically confirm the single-bit error rate of DRAM, something on the order of 1 in 10^13 bits at the time.
Hot spare, hot swap, hot plug...that's how you gotta do it when you have so much hardware on hand that failures need to be planned for rather than prevented.