Transaction weight currently doesn't consider OP codes, it only considers if bytes are part of the witness. Changing that to something more akin to Ethereums gas pricing sounds too complicated to even consider.
I did say per executed OP_CODESEPARATOR, but upon reflection, I agree that we'd like to know the weight without execution. I think counting the number of occurrences of OP_CODESEPARATOR (perhaps at the same time we count OP_CHECKSIG operations?) is a reasonable compromise, and increasing the weight according to my proposed formula based on that count (ideally we'd take OP_IF branches into account).
I would also like to believe that whoever went through the trouble of using OP_CODESEPARATOR reads this list.
I wish this were the case too, but I don't think it is reasonable to assume that (even
maaku isn't subscribed), and I don't even think it is fair to assume such a someone necessarily even speaks English.