They author's point is that Coffeescript is hard to read because it has too many edge cases which are not easily understood under a smalla and consistent set of rules. I have not programmed with CS, but it sounds like a convincing argument.
The point TFA raises is that people who're familiar with CoffeeScript syntax may still make mistakes resulting in completely erroneous but valid code. Yikes. And the inherent nature of whitespace or indentation makes the problem harder to detect by cursory inspection.
I have no strong opinions about CoffeeScript, myself. I just thought this was an interesting distinction, and a rare case where syntax makes a significant difference.
This isn't specific to Lisp, either. I posit that if you took your average C-like and removed indentation from any reasonably sized function or method (one or more loops, one or more if/else), you would have a hard time counting brackets.
I don't think that's completely true. If you're used to algol or C-style, Lisps tend to be hard to read. Concatenative languages even more so. Not necessarily because they're hard to read in and of themselves, but because the "shape" of the code is "all wrong" compared what you're used to, and your usual anchors (keywords, binary operators, infix assingment, ...) are nowhere to be found.
I mean Arabic or Korean are not necessarily harder to read than English, but you've got to know them and how they work before you even recognize patterns in the squiggly lines and little squares.