Wannier90 should still scale cubically with size of the supercell, same as pw.x on Quantum ESPRESSO. Regarding the number of iterations needed, it depends on your system. Assuming you are generating MLWFs (Maximally Localized Wannier Functions) first, the number of iterations required to converge depends on the parameters in your system, like K-Points. Look at Table 1 in this document  for an idea of the relative time-scales for the wannier procedure for bulk silicon. You control the number of iterations using the num_iter tag. But as you might imagine, you can't set this tag to an arbitrarily low value. The correct method should be to set this to a high enough number so that the amount of spread '$\Omega$' in k-space is low. This is indicated by the Im/Re ratio printed in the output.
- Mostofi, Arash A., et al. "wannier90: A tool for obtaining maximally-localised Wannier functions." Computer physics communications 178.9 (2008): 685-699.