Interesting argument. But is it always true that the min. energy config for a large protein has each sub-unit (at least each sub unit on the typical build pathway seen in nature) also in a minimal energy config, by itself?
It doesn't have to. With each added aminoacid peptide is free to reach a new conformation, and the peptide can get locally strained. But each time it doesn't start from nowhere, it starts from the previous conformation, so the number of possible conformations to test in each step is not that large.