"when in reality as we all know, efficiency drops sharply after a point" <- this is simply a rule: every possible task has some granularity past which point it cannot be decomposed, and every network of dependent tasks has a critical path that determines its minimal time expenditure.

Exactly where the "point" at which efficiency drops sharply is certainly project-dependent, but thesash didn't claim that it was not: only that people make the mistake of not realizing that this is even a problem worth understanding and taking into consideration, much less a fundamental one that will likely affect their project.