Hi,
I am not an experto on TDDFT, but seeing your output there is something suspicious. For TD1 you have 4 states occupied (thinking in the 0K electronic structure), and correctly the TDDFT transitions occur from these 4 occupied states to the virtual states in the "Gamma Point". But when you double (or higher) the cell, the excitations occur from states over the occupied states. For example in TD2: 32 occupied states, and the lower citations occur from 38... the same occur for TD3, TD4, TD5.
It seems that the smearing pumps fractions of electrons to virtual states, and the program assumes that virtual-low populated states are occupied states.
If this is the problem, you can solve it by removing the smearing... Si is a semiconductor, and without smearing is possible to compute it electronic structure.
Regards