I'm sorry but I'm having a little trouble visualizing the details of what you've done from the explanation you provided. Presumably you're using element birth and death in the structural calculation to mimic the addition of filler metal (e.g., from a weld rod)? Are you able to ascertain (from post processing) if the temperature calculated from your transient thermal analysis is successfully applied to the structural elements when they are activated with EALIVE? I'm quite certain that body loads such as temperatures are zeroed out when elements are EKILLed. If, for some reason, the temperature body loads are not assigned to the elements once they are activated, then large differences in thermal strain will be seen at the interface between elements that were once "dead" and have been activated and those adjacent to them that have been active all along. That's one thing you might check in post processing... created plots of the thermal strain. Presumably your metal has some strongly temperature dependent material properties around the solidus temperature, so that very hot elements are also very soft (like a liquid). My guess is you're also including a material nonlinearity (plasticity).
I would be tempted to try simulating a weld operation with a Transient Coupled Field analysis system. Coupled field elements support birth and death. Then you could activate filler metal elements in the simulation just as they are added in real life (both thermally and structurally). It would probably take a command object to set this up. I guess this is arc welding? Then you could add electric boundary conditions to newly added filler metal elements as they are deposited to include the effect of Joule heating.