> ... anyhow all calculations during refinement are in fractional coordinates ...
Not necessarily: I can only speak for the RESTRAIN program for restrained refinement that I was involved in developing (the first use of TLS in macromolecular refinement), and at no point were co-ordinates converted to fractional or fractional co-ordinates ever used. All structure factor calculations (and of course the geometric restraints) were done with the original orthogonal co-ordinates read from the PDB file, using the classical structure factor equations (of course more modern programs use FFT for this).
Essentially, you can express the phase factor as exp(2 pi i h.(Sr + t)) = exp(2 pi i (H.R + h.t)) where h is the index vector in integers as read from the MTZ file, r is the fractional co-ordinate vector (not needed in the calculation), S and t represent the real-space symmetry operator, H is the orthogonalised rotated reciprocal lattice vector (H~ = h~SB^-1), and R is the orthogonal co-ordinate vector (R = Br). I think this is the most efficient way of organising the calculation provided the outer loops are over symmetry operators and reflections (so H is calculated once) and the inner loop is over co-ordinates. Of course I accept that using FFT via the electron density is a much more efficient way but I think even then no conversion to fractional is necessary.
Cheers