The same fate would accrue to any human planner who tried making random point mutations to their strategies and waiting 768 iterations of testing to adopt a 3% improvement.

If your engineers are struggling to produce even 1% improvements in your design, and the benefits of even a marginal improvement are sufficiently large, it might well be worth trying such a strategy. The more complex and humanly-incomprehensible the system is, the more likely that such a strategy would yield bonuses that rational analysis couldn't reach, as well.

Rationality is unspeakably powerful, but it's not everything.