I wrote a patch to speed up the cloth modifier. In my environment (i7-10700), the speed was increased by 40% for the benchmark data with approximately 40K vertices.
Since it uses parallelization, I think there is a risk that it might become slower depending on the CPU and data. So I’m interested to see if it will speed up in your environment as well.
Here is the patch (.diff), build binary(.zip) and benchmark data(.blend).
https://drive.google.com/drive/folders/1nl839So5QzP6ukXL30aeAk2n-CCiYIKD?usp=drive_link
commit hash of base code is: 76cf859b3313f3d04d739b8e7b04d259f67bf446