You acknowledge that APL looks complex until you get used to it. So then how can you objectively score the complexity of code written in it? Won't the precise numerical value inevitably depend on the internal state of the individual reading the code? Ignoring this inconvenient fact would seem to lead to fairly useless metrics.
I can see where the notion of scoring overall complexity can be a useful cognitive tool when working alongside people with similar backgrounds to yourself but I don't think there's anything objective about it at all. In fact, I don't think that an objective measure can exist in the general sense.
Note that all I'm arguing here is that any useful software complexity metric will inevitably exhibit a dependence on the individual. I agree that intentional obfuscation adds complexity in an objective sense but would argue that there's no objective way to quantify how much it adds and that any such score will inevitably vary between observers.
I think Code Golf is a relevant example here. Programs of minimal size whose textual representation often looks almost like line noise. They are simple and elegant in the sense that they are small; the meaning per character has been maximized to the best of the author's ability. However, they are also highly complex in the sense that the typical human will find them incredibly difficult to make sense of.