To suggest an odd thing, you might try repeating your tests with the customized set to a single weight — either higher or lower than the default 500 that most programs would use. Berkeley for high-PPI displays is at 450 rather than 500 and looks better for it. (I’m on the Berkeley ship and there are flaws in Iosevka that would see me select Monaco over it, but I’m also a fan of For Science™.)
Do you have a link for "For Science™"?