Notice that OP is defined in terms of the state that S achieved. So it give us a measure of how powerful S is in practice in our model, not some platonic measure of how good S is in general situations. This does not seem a drawback to OP: after all, we want to measure how powerful a system actually is, not how powerful it could be in other circumstances.
I think that if you’re looking for a useful measure of optimization power, you will not want to use actual achievement rather than potential achievement if you want to have a nicely encapsulated concept that doesn’t include properties of the rest of the environment. Clearly the dangerous optimizers are the ones with actual power rather than merely potential power, but I think it’s much more clear to just talk about optimization power as potentially dangerous, given an environment conducive to that optimizer.
I think that if you’re looking for a useful measure of optimization power, you will not want to use actual achievement rather than potential achievement if you want to have a nicely encapsulated concept that doesn’t include properties of the rest of the environment. Clearly the dangerous optimizers are the ones with actual power rather than merely potential power, but I think it’s much more clear to just talk about optimization power as potentially dangerous, given an environment conducive to that optimizer.