I presume that the XG top play you mention includes a move of 6-1 with one of the fives. You can change devious to donk. What is usually happening in positions like this is that XG evaluates the upcoming recube as a take after certain rolls when it is really a drop. And it is probably evaluating recubes after the correct play properly as a drop. The value of the cube mistakes is offsetting the equity lost making a weak checker play play.
If a human makes a move like that, is he cheating? Some time in the 90s, I witnessed Nack do something like 25 minutes of calculations during a match in which he estimated that the value of a cube error he expected to get would more than offset the checker play error he would deliberately make IF AND ONLY IF he rolled precisely a 63 on his following shake. He expected a cube error since the position would be an easy take for money, but was a pass AtS. He had to do the calculations one roll in advance so that his opponent wouldn't catch on to the gambit and stop to think about his action. Eventually, Nack made the obvious move in the outfield, then rolled a 63 instantaneously making the error of moving both checkers to the deuce, recubed receiving the reward of an incorrect take and then won the match rolling the only number (11) which won after the incorrect checker play that would have lost if he had made the correct checker play.
The position was something similar to the one below. What would you think was happening if the best player in the world went into a deep think in a position like this? Never underestimate what the best players in the game are capable of doing.
                  |
| ||||
| XGID=-----A---------A-------bd-:2:1:1:43:3:4:0:11:10 | |||||
| Blue to play 43 | |||||
eXtreme Gammon Version: 2.10