As I understand it, you are using Rietveld refinement only for quantitative
analysis, in which case you may be able to make some headway. If you are using
it for structure refinement, you've got an inapropriate machine and set up and
are on a hiding to nothing.
Given the first of the two assumptions you can try at least 2 things:
1) Forget Rietveld for your whole pattern fitting and use whole pattern
standards. Ref G.Cressey (Nat Hist Mus) March 1996 edition of Powder
Diffraction. Better still give him a ring.
2) Allow the isotropic thermal parameters to vary. Given your data, this should
give you an outlandish negative figure, but a much better fit, which is all
you're after anyway.
Highly recomend opt. 1 if you can get the standards. If you don't fancy either,
the words "shit creak" and "sans paddle" spring to mind.
Happy hunting, Martin