States are shared between the different entry points, so adding entries won’t grow the automaton size significantly (unless large parts of the grammar are completely unreacheable from some entry points, but you should get warnings about that).
The size of the automaton is unrelated to the backend in use, however it is affected by the algorithm (default LR(1) with merging, LALR, canonical).
LR(1) should do well, I don’t think you should worry about that.