It turns out the reason my Amethyst client was so slow was because I had about 40 relays.

I've reduced it to 10 relays now, and I haven’t experienced any slowdowns or crashes. I think I should aim to remove another 5 relays. Not sure what I was thinking when I added 40 in the first place.

ChatGPT agrees that this was the issue and explained it to me in full technical details and suggests I reduce it further to an ideal number of 3.