Yes, we do need to measure quality. And maybe a simple metric wouldn't work, but a skilled programmer should be able to hold two results up side by side and judge for himself which is better.

Controlled experiments with human judges seem like the best answer.