The statistical hadronization model ThermalFist was applied to 17 hadron yields measured in p+p collisions at $\sqrt{s} = 17.3$ GeV.
Recently published yields of $\phi$, $K^\star (892)^0$, $\Xi$, $\overline{\Xi}$,
and $\Xi (\overline{\Xi})^0 (1530)$, measured by the NA61/SHINE Collaboration
were accounted for.
We consistently used the energy-dependent widths of Breit-Wigner mass distributions
of hadronic resonances, as this attempt was found to provide better agreement
with experimental data.
Several variants of the model were applied to the data, including the canonical
treatment of either (i) all the yields or (ii) those with open strangeness,
and the grand canonical approach for the rest.
For the full data set only the latter approach gave moderately reasonable
agreement with data. This result points to the larger volume of strange particles
compared to non-strange ones, weakly supported by femtoscopic analyses
at higher collision energies.
The unjustified removal of the well established experimental $\phi$ meson yield
provided good fit quality for both canonical and strangeness-canonical approaches.