The idea is that polls are modeling a scenario where there is a much smaller amount of mail-in voting, and no significant difference in partisan composition of mail-in vs same day voting (or just not distinguishing between the two at all), and thus little reason that further counting would change the statistical projection. For good reason, as that has historically been the conditions in which presidential elections played out. But growing evidence indicates this will not be the case this year.
It's not, and really can't be known if pollsters should be changing their methodologies to reflect these new degrees of freedom concerning the machinations and consequential effects of mail-in voting; perhaps it won't ultimately matter. But if they don't, and these concerns are even marginally meaningful -- especially as it skews in one direction -- then what the polling is measuring now translates a lot less reliability into the actual tallies on election day and beyond. Frankly i'm surprised someone like Nate Silver hasn't at least wrote some piece with a sensitivity analysis examining the issue.
Wasserman touches on it, with a back-of-the-envelope calculation of an additive 2% effect, which could prove enormous, even if just downballot. And even for those calculations, I think he was using numbers not accounting for the unprecedented volume of ballots likely to occur