Given two "merge-able" chunks of version X,Y and X,Z mergeChunks attempts to give the new resultant chunk version X,A where A is greater than minor version on that shard and fails when that minor version already exists on the other shard.
This always happens when the next minor version is not on the same shard, but there won't be a noticeable failure if X,a happens to be available due to sequence of operations or if a is on the same shard as then we pick the "correct" max.