Now that we have done a lot of editor-interface work in RC 3, giving us the foundation for a dialog to insert and fill template parameters, something like this becomes feasible. However, I don't see it happening in this branch, as I'll discuss in the next paragraph. For now, you should continue using your existing citation templates for the time being. In RC 3, you can make your own "Reference" or "Citation" pack and push it out to the relevant areas. You can recommend that users use it by adding template-rules to the areas. When the user clicks it, they will be prompted for the template parameters you defined (author, work, date published, etc -- again, however you designed your citation templates).
As for a standardized, built-in reference function, there is still an internal debate going on as to how this will be accomplished.
References would use either pre-installed citation functions or templates.
1) The better option appears to be citation functions that can be invoked by BB-Code. In this way we can process the inputs and construct citations programmatically, supporting variable numbers of authors (MLA now supports up to 20 visible authors per citation). We can add some really robust processing and options here, by creating MLA, AMA, et al, libraries and generating the output using built-in functions.
However, this approach becomes problematic when you consider multi-language wikis. The wiki language system is not tied to the forum language system. So if you have only the forum language English, but your wiki has English, French, and German, how do the citation functions translate? Basically, we would not be able to put the phrases in the forum language system. Would Phrases be a new wiki content type?
2) The other option is to use pre-installed templates. Since templates are wiki pages, they can be translated using the wiki language. However, templates could become really complicated (supporting up to 20 authors?!). Currently there is no way really to add multiple values to the same parameter (without having explicit authorFirstName1 through authorLastName20), and linking similar values (constructing: Last Name, First Name) could become complicated in this event.
We may end up using a combination of techniques, but a coherent technical design still needs to be ironed out, along with the potential implications to other existing features, whether they may need to be integrated or would experience breaking behavioral changes.