NiJo,
There is value in what you are suggesting - in fact, I suggested the very same thing to Lawliet before working out my solution. Let me explain why it isn't as great an idea as you might think. After each guess (right or wrong), the candidate list is pruned to only the possible remaining words. If 'q' is guessed and is correct, then the next time around, 'u' will already jump to the top of the list of letters with the most probability to be correct. You don't need to know any special knowledge about letter frequency (alone or in chains) to get this benefit.

In other words, the data you are presented with already takes into consideration only the words that can possible fit given the letters provided and frequency analysis has been done on them. If you have something else in mind though, I am both open and interested in hearing it.