I would think that with various RAID techniques the flash reliability concerns are likely minimized. Add it a little of self-monitoring and log the error rates and you likely can predict when a failure will be significant enough, and then have the device call the mothership and dispatch a technician.
That may be an oversimplification, but it seems like it would be doable with current tech. Heck this was doable 30 years ago when IBM and Amdahl used to do it with their systems.
-----
Invest wisely my friends