All that happens is that they may have to pay more coins instead of less for the interest and capital release. They don't lose money as these coins are free to them and they have a large reserve (which they never even needed to use). The problem when the coin value doesn't rise enough (they can obviously adjust interest rates) is that it will cause inflation.
You're on the right track. When the price of their token starts going down, the company has to pay out more of the token to achieve the desired USD value. This causes the price to drop faster, which causes them to have to pay out even more of the token, and so on in a vicious circle until the token is worth very little and the company runs out of its reserves.
Your use of the term "inflation" is a little confusing, but I guess you're talking here about price inflation rather than monetary inflation. What you call "inflation" is presumably the same as what I mean when I say the price of their token falls.
It's not a problem if at times it isn't deflationary as they don't actually lose any profits. Do you understand now why it's a good model as it prevents the company from dumping the coins as if they did that it means they have to pay more coins for the same amount of daily interest?
Once the price enters its death spiral it will tend to stay in it until it hits the floor. That's why this isn't a good model. They're promising to generate steady USD returns while not having any source of income with which to actually back up that promise other than interest from new suckers buying in. The use of an intermediate token to obfuscate what is happening was clever, but doesn't change the fact that there's no real activity generating the promised returns. While the token price goes up they can pay their promised returns and when it falls they quickly find that they can't.