Well, the sum of all advantages and disadvantages is a scaler utility efficiency value, that you can compare. The best voting method is the one with the lowest Bayesian regret.

It's analogous to arguing about race cars. You may talk about the superior horsepower of car X, and I may talk about the superior aerodynamics of car Y, and some other guy may talk about the lightness of car Z. But talking about these advantages and disadvantages doesn't address the issue. What we should instead do, if we are rational economists, is put the cars through a zillion random trials, with random drivers and road conditions (modeling real life frequencies) and see which car gets the best average time.

That's what Bayesian regret does for voting methods.

You could argue that complexity of the voting method is also an issue, but it just so happens that Score Voting (aka Range Voting) is the second simplest alternative voting method, right after Approval Voting (which itself is actually just a limited form of Score Voting, with a 0-1 "score").

I would like to point out that the Gibbard-Satterthwaite theorem only applies to rank-order method, and does not apply to Score Voting or Approval Voting.
http://rangevoting.org/GibbSat.html

Not to say that Score Voting is immune from tactical voting of course. (Just reacts mildly to it.)

An excellent point. It's nice that Score Voting has this "pleasant surprise" theory, that if all voters give a maximum score to every candidate they like better than their expected outcome, the average voter will be happier than the expected outcome on which he based that vote.
http://rangevoting.org/PleasantSurprise.html