I'll recommend that you do not do Mixed Case if you can avoid it. I know it "looks nicer" but there are a few reasons for my recommendation. The most important reason is documented in one of the MRGs - Autofill Keyword Sets, I believe. But there's a known issue with Cascading Data Sets and mixed case keywords. If you don't ever plan to use Cascading Data Sets that issue becomes moot.
There are two primary instances I will use Mixed Case keywords.
1: The keyword is for descriptive purposes only. For example, if I were routing documents for a review process, I might used Mixed Case for a value like "Routed to Approver", or something else, where that keyword has nothing to do with any process, it's for the user's visual benefit only.
2: The Mixed Case is necessary based on the value of the keyword to the user. For example, if I were setting a keyword that will get added to a notification that sends an email to a third party, I might consider Mixed Case simply to make the email look nicer and less robotic.
Generally speaking though, it's largely subjective, but I feel upper case is the preferred way to go to avoid any possible compatibility issues now and in the future.
