I concur with the assertion that no matter how low the luminosity of the light that reaches the photodetection device it will always be possible to obtain an interference pattern.
One particular striking demonstration of that was in a 1967 experiment by Pfleegor and Mandel: Interference of independent photobeams
The setup used two independent lasers, set up to emit light with the same frequency as close as possible. (Still, laser sources drift in frequency, so the setup had a way of identifying moments when the random frequency drifting brought the two beams very close to each other. At those specific moments measurements were taken.)
The emitted light was attenuated with filters, such that the photomultiplier tube setup that was used registered single photo reception events. (Average time between consecutive events: 2 microseconds)
Any photon reaching the photomultiplier tubes could have originated with either of the two sources. The measurements implied that interference effects were obtained.
More information about this setup is available in a blog post by Chad Orzel, titled The Pfleegor-Mandel Experiment
The outcome of the Pfleegor-Mandel experiment strongly suggest that attempts to account for interference effects in terms of 'a photon interfering with itself' fall short.
Instead the outcome of the Pfleegor-Mandel experiment strongly suggests that in order to account for interference effects properly the setup as a whole must be considered. The proper way to account for interference effects, it appears, entails that for the process of interference the luminosity is immaterial.