Suppose the copy is not a perfect replication. Suppose you can emulate your brain with 90% accuracy cheaply and that each percentage increase in accuracy requires 2X more computing power. This makes the issue not one of "magical thinking" (supposing that a software (not substrate) replica that is 100% accurate is different from the exact same software on a different substrate) to a question of whether a simulation that is 90% accurate is "good enough".

Of course, "good enough" is a vague phrase, so it's necessary to determine how we should evaluate the quality of a replica. I can think of a few off the top of my head (speed of emulation, similarity of behavioral responses to similar situations). It certainly makes for some puzzling philosophy over the nature of identity.