Lindy Effect is not helpful for exploring things with linear scale variance. Its really only a useful heuristic for comparing things that vary by many orders of magnitude. Anyone attempting to use it to compare things or classes of things where there the distribution would be uninteresting on log scale will find it, as you do.

It is however, not bad a heuristic for classifying things in the right order of magnitude. It may seem dumb but the extreme cases are comparing cosmic timescales to geological time scales to evolutionary time scales, societal time scales to lifetimes of humans to lifetimes of insects, all the way to the time scales used in the observations for experimental physics.

I don’t disagree with your results based on the assumptions you’ve posed but I can tell you that if you tried the same analysis with data drawn from fat tailed distributions — the longer and the fatter the tail the better a heuristic it would be. To Taleb’s credit, that means his opinion is right under HIS statistical assumptions.

To that end, it’s usefulness is defined by the problem or data at hand. If had a reason to believe I was in a truly power law setting, I would use it; if didn’t believe that I wouldn’t. All mathematical decision making tools, heuristic or otherwise, are only as good as their assumptions fit the context.