This came up the other day vs. XG. Must've forgotten this one (or never learned it). XG didn't like my play of 25/21 6/4(2) and instead preferred leaving a double shot with 25/23 6/4(2) 13/11! This seems a little wreckless at first glance but makes sense the more you look at it. 6s & 4s are really duped and a lot of the opponent's next rolls are hard to play well. Set up XG's play and then run through all of the oppt's possible rolls and see how often the play you'd make disagrees with XG's choice. A lot of these errors are not the .01 or so variety.
1.  Rollout^{1}  Bar/23 13/11 6/4(2)  eq: +0.169 
 Player: Opponent:  54.08% (G:16.47% B:0.86%) 45.92% (G:12.94% B:0.74%)  Conf.: ± 0.004 (+0.164...+0.173)  [100.0%] Duration: 7 hours 21 minutes 

2.  Rollout^{1}  Bar/23 8/4 6/4  eq: +0.158 (0.010) 
 Player: Opponent:  53.51% (G:16.31% B:0.77%) 46.49% (G:12.41% B:0.59%)  Conf.: ± 0.004 (+0.154...+0.162)  [0.0%] Duration: 6 hours 33 minutes 

3.  Rollout^{2}  Bar/23 24/18*  eq: +0.132 (0.036) 
 Player: Opponent:  52.91% (G:15.21% B:0.76%) 47.09% (G:12.49% B:0.52%)  Conf.: ± 0.011 (+0.121...+0.143)  [0.0%] Duration: 2 hours 22 minutes 

4.  3ply  Bar/23 24/22 6/4(2)  eq: +0.117 (0.052) 
 Player: Opponent:  53.12% (G:15.01% B:0.81%) 46.88% (G:12.22% B:0.52%)  

5.  3ply  Bar/21 6/4(2)  eq: +0.113 (0.056) 
 Player: Opponent:  53.14% (G:14.46% B:0.79%) 46.86% (G:12.16% B:0.56%)  


^{1} 20736 Games rolled with Variance Reduction. Dice Seed: 87281296 Moves: 3ply, cube decisions: XG Roller
^{2} 5184 Games rolled with Variance Reduction. Dice Seed: 87281296 Moves: 3ply, cube decisions: XG Roller

eXtreme Gammon Version: 2.00
For comparison, look at 64S64X22(thx to David Presser for suggesting this). Here the best play is 25/23 8/4 6/4. Maybe the nonstripped 8point is the deciding factor?
1.  Rollout^{1}  Bar/23 8/4 6/4  eq: +0.045 
 Player: Opponent:  50.99% (G:15.16% B:0.69%) 49.01% (G:13.77% B:0.67%)  Conf.: ± 0.003 (+0.042...+0.048)  [100.0%] Duration: 9 hours 23 minutes 

2.  Rollout^{1}  Bar/21 6/4(2)  eq: +0.037 (0.008) 
 Player: Opponent:  51.16% (G:14.05% B:0.76%) 48.84% (G:14.28% B:0.68%)  Conf.: ± 0.003 (+0.034...+0.041)  [0.0%] Duration: 9 hours 48 minutes 

3.  Rollout^{2}  Bar/23 13/11 6/4(2)  eq: +0.029 (0.016) 
 Player: Opponent:  51.10% (G:14.88% B:0.76%) 48.90% (G:14.87% B:0.89%)  Conf.: ± 0.006 (+0.023...+0.035)  [0.0%] Duration: 3 hours 11 minutes 

4.  Rollout^{3}  Bar/23 24/18*  eq: +0.015 (0.030) 
 Player: Opponent:  50.41% (G:13.41% B:0.68%) 49.59% (G:13.73% B:0.54%)  Conf.: ± 0.008 (+0.007...+0.023)  [0.0%] Duration: 1 hour 40 minutes 

5.  2ply  Bar/23 24/22 6/4(2)  eq: +0.019 (0.026) 
 Player: Opponent:  50.48% (G:13.83% B:0.67%) 49.52% (G:13.42% B:0.57%)  


^{1} 31104 Games rolled with Variance Reduction. Dice Seed: 53923039 Moves: 3ply, cube decisions: XG Roller
^{2} 10368 Games rolled with Variance Reduction. Dice Seed: 53923039 Moves: 3ply, cube decisions: XG Roller
^{3} 5184 Games rolled with Variance Reduction. Dice Seed: 53923039 Moves: 3ply, cube decisions: XG Roller

eXtreme Gammon Version: 2.00
Thanks also to David Rockwell for the rollouts.