It's not about test scores, it's about verifiable and quantifiable results. Does dropping a computer in every kid's lap increase the number of hackers? Are we observing more opensource contributions from Peru which can be attributed to OLPC?
Is there an empirical reason to believe that this kind of subsidy is beneficial? This question deserves more than a yawn.
It didn't matter that the chances of running into similar systems in industry were minute unless you ended up coding 6502 for factory automation. It gave a platform for children to experiment on and get used to the idea of coding.
Also, the percentage of children who this inspires doesn't even have to be particularly high for it to have definite beneficial economic effects further down the line and I would be amazed if this kind of effect would be measurable after less than 2 years.