Overclocking without 100% stability is useless.
There is a false notion of stability and how it manifests itself around here. The false logic goes like this: "Prime95 crashes, everything else runs fine! Ergo: the PC is stable in all but Prime95". This is a FALLACY.
Unstability is like a game of chance, with a given seed. The lower the seed, the less errors occur in data processing, the higher vice versa.
Now why does Prime95 crash while other programs do not then? Because Prime95 will:
1) not ignore a single error bit.
2) execute a massive calculating exercise, which will cause as many transistors to be taxed as possible.
The result of this is that Prime95 is a program that can single out (virtually) any trace of instability- *no matter how small the seed*.
The other programs that do not crash immediately seem stable because:
1) some errors are ignored or not directly manifest in human visual inspection.
2) the program does not try to use as many processing resources as possible, which means a lighter (or more specific) load.
So what does that mean to the user? It means that those programs are not suitable stability tests. It does NOT mean that they will never crash or cause problems. Remember the seed of instability- it will cause errors at random intervals, at random transistors which may or may not be currently involved in a crucial operation.
In the long term, the instability, however small, will cause semi-random glitches in those seemingly stable running programs too. Often these may not be bothersome, and some people are willing to accept a glitch once in a while. But then you can never be sure when it *will* finally cause a bigger problem- when it loses your precious saved game or crashes on your Killing Spree.
This trade-off is a personal choice, as long as one realizes that Prime95 and other programs running 'fine' means that they run good enough only to be sure there is no gross instability at hand... and not that they will never crash.