The authors do mention the interesting difference between the analysis using IQ at 20 years old and the analysis using IQ at middle age. They suggest two possibilities to account for this difference:
1) That IQ is, to a degree, a ‘record of bodily insults across the life course’. Apparently IQ scores are inversely correlated with chronic disease, so tend to fall as a person ages and acquires more diseases. The authors tried to account for this possibility by excluding deaths in the first 4 years of follow-up, reasoning that those dying from chronic disease during this period were likely to have the highest amount of comorbidities at study induction, but controlling for this factor didn’t really change their results.
2) That IQ at middle age is more likely to correspond with socioeconomic position than IQ at 20 years old – that is, 20 year olds haven’t had much opportunity to establish or change their socioeconomic position compared to older people.

This study used serveral markers to determine socioeconomic position: income, education, family income, and ‘occupational prestige’ (don’t ask me how they worked out that one). So in theory this analysis would have taken into account ‘stress associated with subordination’ in the occupational prestige measure. Possibly. Health effects of IQ is certainly a murky area of study…