Value of a Feature or Heuristic

[PK]

recently I noticed the following fenomenon, which may make it difficult to measure impact of a single value: there are values that have to maintain correct relation to each other, but not to the values, so to say, from other groups. For example, in my engine changing a knight value, or a bishop value, or a bishop pair value by 5 centipawns produces rating changes that are relatively easy to discern. But setting all the material values to 90% of the original set produced no rating change whatsoever, even though the playing style changed drastically. So depending how you tweak the parameters, material values are either very sensitive (if a ratio between them is changed) or relatively insensitive (if all of them are modified simultaneously in 90%-105% range). So perhaps importance of a change is a wrong thing to measure, and if a parameter is classified as "important", it just means that it is very tightly interrelated with a group of other parameters?