In fact, after thinking twice, I don’t really get the argument. I’ll explain myself. If stealing is a tool for resource management but always makes a pop by abruptly cutting the sound, then sound designers will mostly avoid to use it, but rather use it as a safety net (ideally never triggering).
Eventually, why not leave the trade off to the sound designer, between an instant and hard steal, or a delayed, smoothed steal (with a temporary stealing voice count overload)? It would greatly expand the usefulness of this tool, in my opinion. And, after all, Epic made that choice in UE native sound system.