Thanks for you input, everyone 🙂 As I'm sure is intuitive to many of us, it would appear that the various methods of making ones way through UW is likely going to be the major factor making score prediction uncertain at best.
Having said that, it absolutely makes sense that predictive value could be pulled from UW scores. However, we'd need to gather much more specific data than we have currently. The best I've seen so far was a recent thread which pulled data from a popular IMG forum, but even it had some questionable data points considering it looked at both people who declared their test taking methods as timed, random AND those who didn't mention their method of test-taking.
I'm sure we can all imagine how one might do considerably better on a focused, non-timed study session than random blocks of 46. Of course, the value of UW is definitely in using it as a learning tool, so it would be silly to suggest that anybody use it any other way than what they feel helps them learn best simply for turning it into a diagnostic tool when we already have NBME's for that.