Late to the party again, as usual, but…
One of the things I find myself doing, and reading in just about everybody else's code in most languages, is a prudent sanity-check validation before doing something "dangerous" that profoundly changes the state of the system based on the state of various objects within it. A decidedly non-trivial part of that is usually simple validation of individual property values; is the someFutureDate value actually greater than the value for Now? Should we check that every time we're about to make an important decision based on the value of someFutureDate? What a pain…especially knowing that the one time you forget to do that check often turns out to be the root cause of a critical bug?
ECMAScript 5 has completely rethought how properties on objects are implemented. They can be made read-only, non-enumerable, non-modifiable once set. Perhaps more interestingly, they can now be defined either using a simple value as before, or by defining getters and setters, as is done in several other languages. What this buys you, at the cost of a bit of relocated complexity, is the ability to validate assigned values without the assigning code doing anything other than assigning a value to the property. The domain logic relevant to a specific property of an object can now be coupled more tightly to the property and the details less visible or relevant to outside code, in effect creating a conceptual "mini-class" around a property and its getter/setter logic that is transparent to outside code.
This and several other features of ECMAScript 5 now make it easier to write nice, fine-grained, SOLID code in ECMAScript 5 than was previously possible in any Script dialect. Huge win all around.
Which brings up questions. How could support for property descriptors and the like be added to CoffeeScript? Should they be added? Would support for these underlying ECMAScript features require changes to the CoffeeScript compiler itself, or could it be achieved less intrusively?