Colin, Your point: "I think the point here (probably the one you are making) is that if crystallographers produce a pseudo rigid body motion (or static disorder) model described by TLS parameters then it would make specific predictions of diffuse scatter. These predictions could be used to test the model. In other words data is already there to validate the TLS model and this data is being ignored." is well taken.
This is essentially what was done by Caspar, Clarage et al in their work on insulin and lysozyme crystals, publshed in Nature and Proteins. Nature 332 659-662 1988 Proteins 12 145-157 1992 Diffuse scatter patterns were compared to simulations of correlated motions that were long-range (say, the size of the whole molecule) and short range (say, the size of an amino acid). The TDS data were best fit when the majority of the motion was put in the short-coupled range component with a much more minor component in the whole-molecule component. i.e. it does not seem that TLS models as used to represent most of the B-factor by rigid protein motions are consistent with this TDS data. For that matter, pretty much every type of experimental measurement I know of protein dynamics, including those which probe the normal mode spectrum directly (inelastic neutron scattering work by Cusack) also put most of the motion into high frequency components and deemphasize the low order modes (the TLS components). A review of all this is Prog.Biophys.Molec.Biol. 63 251-276, 1995. From this perspective it is quite surprising that crystallographers are so keen on using TLS models for fitting displacement amplitudes. John Badger