Test cases are no good if they're not used regularly. We've had some test cases sitting in the Mediawiki repository since the beginning of time, and they largely haven't been touched since. They've been allowed to bitrot, to the point of being pretty much useless at this point. The problem is that they were never run regularly (at all), so nobody cared whether they passed or not. The rare exception to this is the parser tests, which have been used very heavily since their inception, and are now integrated to Code Review--any commit to core code causes the parser tests to be run to check for regressions.
If any new tests are to be successful, I think they should be _required_ to be integrated to the code review process. Forced running of tests and clearly displaying the results helps clearly track and identify when regressions occurred, so it's easier to fix them when they happen--which they do. I'd like to see a whole lot of other tests created to cover many aspects of the code, and keeping them as simple and straightforward as possible would be ideal. Making tests easy to write makes developers more likely to write them to include with their patches. If you make the learning curve too hard, developers won't bother, and you've shot yourself in the foot :)
There was also a comment on the talk page from Michael Dale mentioning that this should integrate with Code Review. This is definitely a good idea, and I'll try to work it into the plan.
As for the tests, simple tests seem fairly easy to create, as you can use Selenium IDE to record actions, then output the script in a few different languages (I'm assuming we'll use PHP). I think the hardest part is going to be keeping the tests up to date with the code.
Respectfully,
Ryan Lane