One could use a code-challenge scoring rule, that combines some measure of the inherent golfiness of the language with the code length on this specific challenge.
For instance, one might say that the inherent golfiness of a language is the (geometric) mean length of the shortest answers to each of a selected group of popular challenges.
Then, the score of a submission could be the ratio of the length of that submission to the inherent golfiness of the language used.
For instance, if the inherent golfiness of Pyth was 10 bytes, while the inherent golfiness of Python was 100 bytes, then a 50 byte solution in Python to your challenge would beat a 6 byte solution in Pyth.
Scored submissions would only be scored this way in languages where enough solutions had been submitted to the selected group of popular challenges before the new challenge was posted to establish an inherent golfiness for the language. Submissions in other languages would be allowed, but would be treated as having inherent golfiness 1 byte, and so would almost certainly not win.
This would be a better way to compare golfs across languages than the current method.