From the very basic relation between MW frequency and resonance magnetic field (see mt2mhz / mhz2mt functions), one is proportional to the other.
I would therefore expect that line widths in MHz (for frequency-sweep experiments) and mT (for field-sweep experiments) would be interconvertible with the above functions.
Is that, however, rigorously correct? Is there some finer factor that has to be considered?
The more fundamental property is the frequency-domain line broadening. In EasySpin, it is assumed to be Gaussian. Whenever you use any of the strain broadenings, this gets used.
EasySpin uses field-domain linewidths (Sys.lwpp and Sys.lw) just for convolutional broadening. Therefore, every orientation gets the same broadening, even if the g tensor is very anisotropic. There is no physics behind these convolutional broadenings, i.e. no mechanism is assumed. It's purely phenomenological.
In general, the field-domain line shape resulting from a Gaussian frequency-domain broadening is non-Gaussian and asymmetric, if the energy level dependence on the magnetic field is non-linear (which it is very often).
A good read regarding this topic is Pilbrow's book on Transition Metal EPR.