That is exactly the case I was talking about. There is nothing you can do about the time it takes to search the best move when you only split at the root. But most other moves would be below it, and these could then be quickly refuted even without knowing the exact score of the first move. So it is very possible that 3 other threads can quickly prove all other moves fail low compared to the aspiration alpha, while the first thread is doing the first move, so that in the end your search time is reduced to the duration of the latter.lucasart wrote:You're taking an edge case. What you should look at is the general case instead. In the general case, the first move is best, and the score is within the aspiration window's bounds.
Even when some other moves score close to the best move, when you do LMR you could postpone their full-depth search until you have refuted all other moves at LMR depth. By that time the result of the best move might be in, and they could be searched with the same alpha that a single-threaded search would also have used (except now 4 at a time).