Those ratios probably vary widely on the real world. Many are probably near 80%/20%, while a few may be around 99.9%/0.1% or 50%/50%.

As you get more data points, your generalization fails more and more often, and some failures are severe, adding more planning risk than everything else added.

Or, in other words, the rule works most of the time. And you can rely on it at all.

there is a common conceit that if you repeat a word often enough it stops making sense, but in this case the statements are, as often is the case with percentages regarding tech, meant to sound meaningful but are probably not as clear as they could be.