Hmm, we're talking in the same general space but miscommunicating on details, so this discussion may need to wait until you guys can provide specifics of what you're working on. But let me try to clear up a couple points.

The thing is, is that really what people who are building applications need to figure out? Yes, they probably need access to changing these properties

You answered your own question. Yes, they need this, along with a lot of other things. That complexity is intrinsic to making a web app and therefore is hard to completely abstract away.

Of course I do not exaggerate the importance of div borders. My point was that even the sexp your colleague cited as a counterexample contains a lot of information - not just browser nastiness or crud, but useful information, which any model you try to replace HTML/CSS/JS with will still need to represent.

It isn't so much about compressing this

When I said "compression" I didn't mean it literally, as in gzip, but metaphorically. A good abstraction is a kind of logical compression inasmuch as it replaces a complex, detailed model with a simpler one that contains the same information. It does this by throwing out non-essential details (like duplication). Now, if your goal is to build a general abstraction ABC on top of HTML/CSS/JS, ABC must be much simpler in order to justify its learning curve and the fact that it's nonstandard. (By "nonstandard" I simply mean unfamiliar.) That means you're going to have to find a lot of "noise" to throw out. My skepticism is based on the fact that I don't see a lot of such noise in the general case. A thing can be messy and yet still difficult to abstract over. But I'm not saying it can't be done, and good for you if you can do it.

The approach we take with Parenscript is more modest, I think. We're not building universal abstractions for web apps, just for our own. That means we can abstract away from anything we don't need, whether or not it's needed by others. That and the fact that Lisp gives us a homogeneous way of representing all the HTML/CSS/JS pieces are two very big wins. Is it enough to get non-Lisp-geeks to use it? Probably not. But I don't think that's why most hackers take up something like Lisp in the first place.