The tally is not normalized by volume, so the units are something like [neutrons-cm/source]. This can be understood quite easily: in a collision estimator tally, OpenMC accumulates 1/Sigma_t which has units of cm, and similarly for a tracklength tally, OpenMC accumulates the tracklength in cm. When you divide by volume, you get [neutrons/cm^2-source]. The normalization factor Pnu/(Qk) gives you [J/secneutrons/fission/(J/fissionneutrons/source)] = [source/sec] so that when you multiply the flux by the normalization factor, you get [neutrons/cm^2-sec].