My response yesterday is not 100% accurate. I mistakenly presumed you were referring to a problem in server-side JSON parsing which was not fully consistent with the change to server-side JSON creation, introduced in 2.2.3, and this has already been fixed in our code.
However, the basic fact that server-side JSON creation has indeed changed in 2.2.3, and that it is now inconsistent with the client-side implementation, as you demonstrate.
Note that while inconsistent, both versions are considered valid, since Unicode character escaping is permitted but not required - see for instance http://stackoverflow.com/a/4908960.
The annoying fact is that the current implementation is based on standard serialization provided by the underlying platform (Java/Browser), which replaced the previous, custom implementation we had on the server-side.
Having said that, we shall look into providing some form of backwards compatibility, providing the seemingly consistent result.
For best results, use the Firefox browser..