Feedback scores should be based on simple +1/-1 semantics are by far the most prevalent among consumer sites(eBay/Amazon/YouTube/Reddit/StackOverflow) and seem effective. However, I think the score displayed should be twofold: all-time cumulative, as well as weighted. With the rise of the "trending" classification system, a user should be able to quickly discern whether a score is recent.
Another alternative in this vein would be to have "rounds" of scoring. That is, cumulative score is permanent, and while it can be impressive, in all cases it appears to stagnate. But scores within the round are more dynamic. The thrill of a new round, experiments with strategies, and end-run pushes for big finishes are all there.
One disadvantage of both arrangements is the potential for accidental alienation of experienced users. Precious few users use a site forever, but overaccentuating the passage of time can certainly accelerate losing the old guard. Every site is different, and careful assessment of this risk is critical to preventing the loss of institutional knowledge and parties willing to maintain foundational artifacts of an internet property.