In terms of computation efficiency, magnetoelastic field itself is not a problem. Demag remains our bottleneck. When you need time-variance in the MEL field, you need some tcl scripting or reloading of strain every so often during the simulation. The tcl approach can be very slow, since it is evaluated at every simulation step. YY_TransformStageMEL class takes the second approach and interpolates strain/displacement at given two moments. This is much faster, depending on your choice of strain-reload frequency.

In terms of ability to handle complex structures, I think Phys. Rev. B 90, 140405(R) is a good demonstration. You need to get the strain or displacement distribution by some other means (e.g., FEA).