The Pentium overclocking ventures clearly proved that overclocking is, at best, risky.

I once had responsibility for proving just that, and it took less than a day and 60 identical systems to prove it. Once they were up to temperature, monitoring showed that as many as 1/3 of the machines, all of which were exactly identical in their equipment and configuration, behaved differently from the others when all were subjected to the same clock setup and fed the same keyboard and mouse inputs. Even fairly elementary DOS-based operations "fell apart" even when the clock setting was only one step above the "official" rating, once the systems were stressed and had been brought up to maximum allowable temperature (25C) which seldom is the maximum to which commercial quality systems are, in fact, driven. Fortunately, I didn't have to prove that they always failed, just that they occasionally failed.

Just because two out of three MCU's can exceed their specified clock rate limit for a short time and on a limited instruction and data set doesn't mean that they can properly execute every combination of instructions in every possible location in their memory, both internal and external, on every possible combination of data. It certainly doesn't mean that 20 out of 30 will do that, and absolutely doesn't guarantee that 200 out of 300, coming from different lots, will work in that way.

Go ahead if you're willing to gamble, but please let me know if you ever design a commercial product that way, so I can avoid it, and every other product from that mfg.

RE