I performed the tests you suggested (see attached table). When I use a larger error tolerance (1e-5) for Ewald and PPPM the resulting pressure is larger than what I got for a tolerance of 1e-12. The resulting pressure is still much better than the analogous MSM calculation though. If I add the “kspace_modify cutoff/adjust no pressure/scalar no” command for the MSM calculation with a tolerance factor of 1e-5 I get very good results with the MSM solver (independent of initial configuration), comparable to Ewald with a tolerance factor of 1e-5. I do not quite understand why the addition of the “kspace_modify …” line achieves results which are so much better than without, I do think that the problem is—as you mentioned—the incompressibility of the crystal combined with the tolerance factor of 1e-5, because with all solvers the pressure gets better with smaller tolerance.