I agree with the sentiment, but only for comparisons of the same language. There's no point comparing C to Golfscript -- Golfscript will pretty much always win. But it would be nice if the better-scoring of the C answers were higher. As I see it, code golf is a per-language competition.
There's an unfortunate trend where mediocre golfs written soon after the post float quickly to the top, getting visibility and thus more upvotes and more visibility. On the other hand, a golf posted after the initial rush that took time to squeeze out characters and improve on existing golfs risks lingering at the bottom, ignored.
Sometimes, the submission floats up, but other times, it never gets the initial jolt, perhaps because nobody scrolls down to see it, or because their attentions have moved on. It's frustrating to spend time on a laborious and well-documented golf, only to see it ignored in favor of submissions that took less effort and show less ingenuity.
I know that upvotes are meaningless Internet points. But, they carry with them attention and comments, and an implicit acknowledgement of a job well done. It would be nice for them to consistently reward good golfing.
This is not to say that upvotes should only be based on code length. Method, creativity, and language choice can also inspire someone to upvote, even on objective challenges. But as is, I believe submitting early is at least as effective as submitting well-golfed code for getting upvotes, and that's no good. (This probably doesn't apply for languages like Golfscript, which float up regardless, perhaps because voters are endlessly amazed to see random-looking strings of characters).
I make a conscious effort to upvote shorter solutions (language-adjusted), and sometimes that means not upvoting a longer one of the same language, even if I still like it. But, I don't think the solution is to get others to do the same.
I think a technological solution is needed. Something like a sort-by-score with a by-language filter. This would require new tech, and I don't know how possible it is. Ideally, the score would be assigned by the question-asker or perhaps voted on by the community to resolve disagreements.