The "problem" appeared when i tried to translate the following sentence from Tatoeba.
"My back hurts."
If we assume that back is {bekpi} (cuz Robin needed a gismu for it)
then we have three options.
1. mi cortu lo bekpi
2. cortu lo bekpi be mi
3. mi cortu lo bekpi be mi
The third solution is verbose and therefore doesn't reflect relations between sumti in a nice way.
I guess Lojban is just unable to express this in a more concise way.
May be
4. bekpi je selcortu mi ?
I wish I could bind {bekpi} and {selcortu} with {du} but I can't.
I created a semi-serious solution to this problem when it first occurred to me too. I propose using a property + an indirect question as a verbose albeit consistent system: {.i mi cortu lo ka [makau] bekpi [ce'u]}.
My rationale for disapproving of concrete sumti in this case is the same as my rationale for disapproving of events in kakne2: you should be able to use any concrete sumti (or event in the case of kakne2) but you can't because bullshit. (Indeed {mi kakne lo nu do citka lo plise} is nonsense.)
This extends to any case where you have a concrete sumti with the restriction that it is intrinsically tied to another sumti.
e.g. {mi bajra fi lo ka makau jubme ce'u}
Naturally, we can use tanru (and jvajvo!) to make these formulas shorter. {mi bekpi cortu}.
.i mi'e la tsani mu'o