Are there empirical facts that can answer that question? It looks like a question about preferences to me, which are difficult to measure.
I think you're right that many of the relevant empirical facts will be about your preferences. At risk of repeating myself, though, there are other facts that matter, like whether ems are conscious, how much it costs to prevent torture, and what better things we could be directing our efforts towards.
To partially answer your question ("how much effort is it worth to prevent the torture of ems?"): I sure do want torture to not happen, unless I'm hugely wrong about my preferences. So if preventing em torture turns out to not be worth a lot of eff...
When I was reading The Seven Biggest Dick Moves in the History of Gaming, I was struck by the number of people who are strongly motivated to cause misery to others [1], apparently for its own sake. I think the default assumption here is that the primary risk to ems is from errors in programming an AI, but cruelty from other ems, from silicon minds closely based on humans but not ems (is there a convenient term for this?) and from just plain organic humans strikes me as extremely likely.
We're talking about a species where a significant number of people feel better when they torture Sims. I don't think torturing Sims is of any moral importance, but it serves as an indicator about what people like to do. I also wonder how good a simulation has to be before torturing it does matter.
I find it hard to imagine a system where it's easy to upload people which has security so good that torturing copies wouldn't be feasible, but maybe I'm missing something.
[1] The article was also very funny. I point this out only because I feel a possibly excessive need to reassure readers that I have normal reactions.