Perplexed:

The Salthouse paper states that the Z-scores there are relative to the entire sample (ages 20 - 60). But I'm pretty sure that the OP's graph uses Z-scores relative to the variation within an age cohort.

The poster should definitely clarify that. The current presentation is unclear.

Otherwise, they would be too steep to be believable - almost mathematically impossible considering that the sample runs out to age 80 and shows a drop of about 2.0 Z units

I don't find it that implausible. Anecdotally, I've met people whose cognitive skills appeared horribly diminished already in their seventies. (Don't forget that the median age of death in the developed world is roughly in the mid-seventies for men and late seventies for women.)

Even if cognitive decline can be greatly ameliorated by persistent intellectual effort, most of the population clearly won't benefit from this, which is good to have in mind if one has seen an anecdotal sample of exceptionally intellectual people.

ETA: The Flynn effect may account for about a third of the apparent drop in cognitive ability - that is, today's old-folks started off less cognitively skilled than today's youngsters, so their decline with age is not as steep as it looks.

As in all other things related to measuring cognitive ability, the Flynn effect tends to make a confused mess of the whole situation just when one starts to come up with neat and plausible theories. It's very hard to tell how and to what extent it has bearing on these results.