> I have done an exhaustive search and none of the 6 situations of 18 moves
> could be reduced to 17 moves (within the group of <R2,L2,F2,B2,U,D>).
> For the case you want to verify, one of them is:
> L2 U R2 B2 U2 B2 L2 D2 L2 F2 D' L2 B2 F2 L2 F2 U' D
Of course I verified it ;-). This one does indeed kill Kociemba's algorithm.
On a fast processor (65 MHz SPARC) with a larger limit database than Kociemba
is using himself (the database is about 5 MByte for the second phase), it took
3 hours 15 minutes to find a minimal solution. Of 18 moves.