This is the crucial issue, yes.
However, one can think of a level-assignment process that runs as follows.
First, you assign a general level to any abstractl result (killing, healing, scrying etc.) regardless of the specific effect used to reach it, possibly with modifiers due to Arts. Say, killing a person without fuss at good range should be level 40, maybe 35 if using Perdo, increased or decreased depending on range (kissing the target deserves -5 to -10), restrictions on use (only works in a forest? -10 to -15), quickness, side effects etc.
Then, the player describes a concrete effect, and the troupe thinks of all its possible abstract uses. E.g. if you can move any boulder lying aroud, you can use it both to kill and to destroy stuff (though probably not in confined environments, and you need the boulder around to start with ... but then with such an effect you can probably always carry one around), as well as to build fortifications and other structures, to fly (on top of it) etc.
Finally, the troupe assigns to said effect the level based on the highest abstract use it can be put to, increased by significant flexibility (say, something that can be used to kill and fly and build/alter structures) and possibly decreased if its appropriately mythic (having mist rise slowly from the ground is more mythic than having a cube of mist 10 paces on each side suddenly materialize).
This does not create the problem of rock-creation being of a different level depending on whether they are created on top of someone's head. It solves game balance. It also solves what I call the "unmythic problem", which is in my view not so much a game balance issue, but an aesthetic issue:
However, this approach by necessity takes away that component of fun of ArM5, namely to find the easiest way to achieve a specific abstract result (e.g. killing an opponent) with the "cheapest" possible magic. It's a tradeoff.