The issues I've personally experienced have been with impact, for bugs outside of the very traditional XSS/SQLI/RCE. I've gotten things along the lines of "yes, our _______ is seriously broken, but it's not/barely a security issue," with an explanation that stretches plausibility. Maybe I'm full of crap, maybe they are.
I'm sure those running bounty programs would have all sorts of folks contesting things that aren't actually real bugs. I think the only real good solution is increased visibility on all sides. That way each of our technical arguments can stand on their merits, whoever is full of crap can get called on it, and others looking for bugs can choose where to invest their time (glossing over that solution missing a bunch of thorny implementation details, I'm sure).
Whatever the scale is, companies have no incentive to avoid paying them, because even at the high end of the scale the amounts are immaterial. Remember, we're talking about H1 bounties here, not the Apple and Google platform bounties, which are totally different animals.
The real risk companies that run bounties face is that their programs won't generate any real bugs at all, but will absorb costs from both the platform and all the nonsense bugs they have to triage. New serious bounties are good news, not bad news, for most bounty programs.
(I've managed several, continuing until recently; before I did, I went around talking to people who ran them to get the lay of the land. I'm pretty confident in my answers here.)