As you can imagine, it is extremely difficult to "generalize" what practical significance is as it is completely dependent on the situation and the response variables (Y's). My guess is the 10% of the spec (which, unfortunately, is often an arbitrary value) or of the distribution of the data (6s) value is an artifact of the 10% rule of thumb for measurements system components of variation of the overall variability. I ALWAYS suggest you consider the practical significance of the data before performing any statistical analysis. The problem is how to "force" the users to do this? When explaining this concept to others, I always think about:
1. What is the smallest increment of change of the Y you care about or would be of interest (engineering or scientific)?
2. If you could move the Y, what is the smallest increment you would be willing to spend resources to move it?
Practical>Graphical>Quantitative is the order of analysis.
"All models are wrong, some are useful" G.E.P. Box
