It's kind of weird because more machines working together does mean more work gets done. Coin flips and weighted coin flips are totally different things. Any biases weights towards reality push you closer to reality when you use them.

New models keep being able to use more and more agents on longer time frames. Your hypothesis doesn't look like what we're measuring.