Another key point to bear in mind is that if one is asking for the "hottest" star, then presumably it is "effective temperature" that is referenced. Effective temperature is related to the luminosity L and the radius R from which the light emerges (not the static surface of the star, R can be well out in a dense stellar wind) by the Stefan-Boltzmann law, which says T_eff ~ L^(1/4) / R^(1/2). Normally for main-sequence stars, whose R rises much more slowly than L, that means the T_eff is higher at higher L. But really high L stars have dense radiatively-driven winds, and eventually the wind could get so dense that the mass-loss rate from the star might be proportional to L. If so, the radius R from which the light emerges (where the optical depth is near unity) is also proportional to L, so then T_eff ~ L^(1/4) / L^(1/2) ~ L^(-1/4).
This says we should expect T_eff to reach a peak and then drop if L is further increased. The peak T_eff should be when the winds are getting so dense that the mass-loss rate becomes proportional to L (when a significant amount of L is being used to lift the mass), and that is more or less also where the WNh spectrum replaces the O spectrum (something like what is trying to be the O1 spectrum). So the kinds of T_eff already referred to, above 50,000 K or so, is going to be the highest T_eff possible for a core hydrogen burning star. After that, even higher L will simply mean that you see a much larger star with a lower T_eff.
Incidentally, this probably points to a flaw in the Bromm et al. paper cited above. That paper is really about Pop III stars, so zero metallicity and not much in the way of stellar winds also. So when they talk about T_eff, they appear to use an expression that involves L and the static radius, but that's not the T_eff that matters for observing stars, or for their UV flux. The latter should use the R from which the light emerges, which can be in the wind. The distinction is not important for the pop III conclusions of their paper, but it does matter for the Pop I results shown in Figure 1. That is probably not the correct T_eff, so I doubt T_eff really can get up to 65,000 K for Pop I stars.