I don't necessarily agree with this line of argument. But people will keep making it, so it's worth formulating a response to it if you want to win the debate. (Not that rational discourse wins any debates these days.)
In addition to that opportunity cost concern, there's knock-on effects, like you touched on, mustpax. Letting a killer live saves one human life now, but a policy of letting killers live may end far more human lives in the future. Personally, I don't think the data supports that, but if it did that's a choice we'd have to make.
The magnitude of "overall harm" is not an issue.
I simply adopted the axiom of "nothing is more important than a single human life" and came up with one approach to make sure that axiom is never violated. There may be better approaches.
Agreement on this axiom is a necessary first step. I'm not sure we have it. What a shame.
In real life, there is no escape from balancing potential risks (in this case, of innocents being killed) and benefits.
If you've ever worked in public safety or law enforcement, you'd realize how uninformed that opinion is.
Law enforcement officers don't carry weapons to use them, they carry them as deterrent. Good law enforcement officers regard the use of deadly force as either a last resort measure or just bad police work.
If it's too much trouble to modify existing systems and instutions, the just use this handy formula for binary decision making:
Dilemma solved.