Sounds nice. I especially welcome the variable mechanism for colours so that I don't have to write the hex value of today's favourite color (or its highlighted or shaded variant) everywhere I use it.
I'm not sure yet if I'm happy with the lazily evaluated variables, though. In makefiles, I tend to prefer := assignment (eager evaluation) to = assignment (lazy evaluation) when possible to avoid surprises.
It doesn't seem like it's about getting a better result. As far as I can see it's allowing people to get a result in less time.
Imagine if CSS made you style every element individually, so you had to specify the full path through the DOM before applying each style. It's still the same language because you could get the same results - it's just a lot more typing.
It seems to be a similar thing happening with CSS frameworks, although there's less inefficiency to remove than with my exaggerated example.
Adding variables and arithmetic to the mix also has the potential to make things more maintainable if used correctly - if you're setting the same number multiple times or setting a number as the result of a calculation an expert CSS coder will have to remember to update everything (and/or) redo the calcuations every time they want to make a change.
It's a medium sized win but I'm willing to take a look.