I suspect, but don't have any proof, that it is the physical difficulty of getting both ends ligated to a single strand - think of it like this:
Your 10 kb is essentially a long string, ligate one end to your vector, now try to get the loose ends of that now ~13 kb string in proximity to the same. All in a tube with (assuming nanomolar concentrations) approx 1011 other molecules all trying to do the same. Chances are that the loose ends will encounter another molecule and end up ligating to those rather than the end that will result in a closed circular plasmid.
In contrast, with shorter molecules, having the insert and vector ligated at one end should constrain the number of potential interactions by relative proximity, making it more likely that the loose ends will interact and ligate to form the circular plasmid.
Now, I am sure that this isn't the complete answer, as there will be DNA topology, reaction conditions, etc. to take into account, but I think it is a good approximation of the problem.