Is that so unreasonable? 90% availability still means that any given machine is down 1 day every 10 days. That's a potentially prohibitive cost to a company.
There are military systems that specified for >98% availability. Our IT network is required to have availability of around 98% as well.
The key to that is the ability to diagnose the problem quickly and repair quickly. This might entail having all most plausible spares on-site. Additionally, this requires your diagnosis tools to provide an answer in less than a few minutes and the ability to perform the repair in less than a few hours, assuming availability of spares.
Availability defined as: MTBF/(MTBF+MTTR+MLDT), where:
MTBF --> mean time between failures
MTTR --> mean time to repair
MLDT --> mean logistical delay time (parts availability)
A high MTBF is obviously desirable, but you can see that MTTR and MLDT can eat your lunch if it's not well controlled. Let MTBF=600hr, MLDT=5day, MTTR=3hr, Avail=83%. If MLDT is reduced to 1day, Avail=95%
TTFN