I recall from someone saying that 'max and min' aren't the best measures of cooling performance. Rather, when you average things out that's when things get interested. Could you re-analyze your data by average load / idle results?

I believe SKYMTL said it. And I think it has grains of truth.