My personal experience tells me that a bit flip is very unlikely to cause a system to fail before a bug or normal termination turns it down. My intuition tells

My personal experience tells me that a bit flip is very unlikely to cause a system to fail before a bug or normal termination turns it down. My intuition tells me this is due to shorter and shorter individual program life span due to expansion of client/server architecture, higher frequency of bugs due to expansion of untyped/unstructured data, and the fact that most data manipulated by a program is not initialized/used anyway. Only in very big clusters have I ever heard of memory failure being noticable, but even there I've never seen memory corruption cited in a post mortem. Having to deal with bugs and other hardware defects forces one to work around lack of data integrity anyway.