My concern about using EKILL to deactivate elements whose temperature exceed 2500 C is that when elements are deactivated, the element loads applied to them are "zeroed out of the load vector". I'm pretty sure this includes surface heat flux loads. So if you deactivate elements on the surface where the heat flux is applied, some (possibly much or even all) of the net heat flux applied to the surface will no longer be applied to the model. This may be inconsistent with what actually happens in the physical system.
Have you defined an enthalpy material property? We use this to represent the latent heat associated with phase change. When one defines enthalpy, density and specific heat material properties are no longer used (the enthalpy material property replaces their combined effect of storing heat).
You could have two step (actually steep ramp) changes in enthalpy with temperature: one associated with transition from solid to liquid at the melting temperature and one associated with the transition from liquid to gas at the vaporization temperature. If you do this, your analysis might not predict such high temperatures, and if you don't deactivate elements, the net heat applied to the model will not be artificially reduced.
A couple of tips from a very old training course notes that you might keep in mind...
One other thing is that in ANSYS, enthalpy is defined in units of Joule/m^3, whereas outside of the ANSYS world, I always see latent heat defined in units of Joule/kg. As simple as it may seem, it can be easy to make a mistake in the conversion.
I hope this is helpful.