If you have a model which is in error because of an overall rotation and translation, or even if only one domain has such an error, none of the refinement packages will be able to correct this automatically. To correct this error a concerted shift of a large number of atoms must be made. The indication that such a shift is required in located in the off-diagonal elements of the normal matrix, which has been discarded. For this problem to be corrected you will have to resort to rigid body refinement.
In rigid body refinement the molecule is defined to contain one or more groups within which the atoms cannot move relative to one another. Basically the parameters of the model are recast in a form in which there no longer are correlations between the parameters.
While it is unlikely that errors of this type will arise when solving a problem with the MIR method it is quite common in MR and even Molecular Substitution (Isomorphous inhibitor or mutant structures). Refinement in these latter cases should always be started with overall rigid body refinement and refinement with each domain as a separate body. There have been a number of cases where the refinement has ``hung up'' in the mid to upper 20's (percent R-factor) where the problem was eventually traced to a small unguessed domain shift.