How much does computing power matter for a Xiangqi engine?
In real games, the advantage of more computing power is hard to see. Real games start from balanced positions that are very forgiving, and there are opening books that dissect the starting position in great depth, so the draw rate is extremely high. Under normal conditions, even a several-hundred-core monster against an ordinary machine, with all other variables roughly equal, still draws at least 97% of the time.
That does not mean computing power is useless; it depends on what you need. To put it another way: the high-end machine might win 20 games and lose 0 out of 1000, while the low-end machine wins 5 and loses 5 out of 1000 (these numbers are made up, not real data, though not far off). If you think that is useless, it is useless; if you think it is useful, it is useful. Also, the faster the time control, the more computing power shows, but unless it is a one- or two-minute sudden-death game, it is still hard to see.
From a testing point of view, however, the difference between strong and weak computing power is huge. To tell quickly which engine is stronger, testers use paired games from unbalanced positions, for example high-advantage openings with a 50% draw rate, where the two engines play both colors in every opening. Under such unbalanced paired games, the strength gain from computing power is substantial; for example, 2 threads at 5 minutes + 3 seconds against 1 thread can reach a win rate above 55%.