Interesting. It's more than a couple of lines of code to change for Jython (I tried), but now that we require Java 7, we should be able to implement identical ordered behavior to PyPy for dict, __dict__, set, and frozenset using ConcurrentSkipListMap instead of our current usage of ConcurrentHashMap.
How? A skip list is a sorted collection--kept in the natural order of the keys, not the insertion order. Of course you can transform each key into a (next_index(), key) pair at insert time to keep insertion order instead of natural sorting order, but then you lose the logarithmic lookup time when searching just by key.
And even if I'm missing something stupid and there is a way to make this work, you're still only going to get logarithmic performance, not linear. And you're probably going to change the requirement on dict keys from hashable to fully ordered (which, on top of being different, is substantially harder to detect at runtime, and also doesn't map cleanly to immutable among Python's standard data types). So, I don't think it would be a good idea even if it were possible.
(There's a reason that Java, as well as C++ and other languages, provides sets and maps based on both hash tables and sorted logarithmic structures--because they're not interchangeable.)
(Other collections like defaultdict and weakref collections use Google Guava's MapMaker, which doesn't have an ordered option.)
I think this is a great idea from a user experience perspective.