It also depends on the fundamental resilience of the system, whether or not single failures compound, whether or not the fail-safes themselves have failures or faults within them, and how personnel (and management) respond in the event of failure.
Known, simple, redundant, and stable systems which tend to return to modes of stability, which don't tend to experience runaway failure modes, and whose staffs are trained in known (and unknown) failure modes, tend to work well.
Unknown designs (they or staff are new, they're poorly documented, they're acquired from vendors or through organizational acquisition, etc.), whose staff aren't trained in normal and abnormal operations, which do tend to go into runaway failure modes, whose safety or management systems themselves have (known or unknown) bugs, etc., all tend to compound failure modes.
I've had direct experience of this at several levels myself. More frighteningly, I've interviewed senior management of a nuclear facility who candidly admitted that it was poorly managed.
Realize that a 4GW nuclear power plant is producing about $360,000 worth of retail electricity ($0.09/kWh) per hour, and that downtime costs over a million dollars every three hours. Keeping that plant online and operational has a very high priority -- sometimes to the point of cutting corners to do so if short-term objectives may be met at the cost of long-term sustainability.
