I think there must be something fundementally wrong with
your X-PLOR calculations. A good test would be to plug
your REFMAC refined model into X-PLOR and you should
come out with reasonable R-factors (If you apply the X-PLOR
anisotropic scaling and bulk solvent correction the results
should be extremely comparable, but even without these they
should not be so hugely different).
In our trials with Maximum Likelihood implemented in X-PLOR 98.0 we
found a somewhat greater impact on the performance of
minimization-based protocols than torsion angle dynamics but
even if you are using torsion angle dynamics without ML
you should still have a good radius of convergence, especially
with your high resolution data.
Product Development Manager, Macromolecular Crystallography