I'm not 100% sure exactly how they differ so I have tried all of them. I think the Oracle/Oracle 4 try to find the best fit to individual populations (lower numbrs following the population indicate closer fit) and the Oracle-x works out the most likely distribution among populations. The Pct. 1 and 2 are a bit of a mystery to me but I found that Pct. 2 fit best.
The folks on the FTDNA forums might know better. I'll post your question there.