I've observed that error rates are not the same for every machine - there are very bad machines, which usually have some way of triggering enough errors systema

I've observed that error rates are not the same for every machine - there are very bad machines, which usually have some way of triggering enough errors systematically that they end up getting someone's attention, then there are bad machines which have memory errors or are observed to report errors or crash occasionally, and then a bunch of good machines which on the surface appear to work flawlessly. Sometimes what makes a machine bad is the memory itself, other times it's the hardware or the configuration that the hardware settles on. The statistics may be skewed by the uneven distribution of "bad machines" amongst good ones, especially if you take into account hardware in the hands of end users rather than servers. It gives a new meaning to "works on my machine."