It takes something that is, in human terms, relatively simple and makes it so convoluted that it's hard to even follow.
I read the Lightning paper and found it simple enough to understand (conceptually at least) how the channels are opened, updated, closed, and penalized. Of course the actual implementation is a more complicated and nuanced than what the paper covers.
The same volume without lightning would require blocks that were terabytes, which is obviously unworkable at the current state of technology.
A "black swan" event like a major lightning hub going down may force more channel closings than the network (as currently implemented) has time to process before time-out.
The LN simply does not work if the base layer is congested.