It’s not so that “transaction fees will go up”, at least in any limited sense. If the limit is reached, that produces a backlog of transactions. And as long as the number of transactions that wish to go through is steadily greater than the number that CAN go through, this backlog will grow.
Meaning that transaction fees will go up. But not by 2 cents or some other limited amount. As long as the number of desired transactions is greater than the number of possible transactions, the fees will continue to go up ad nauseam, while users that don’t pay these ever-rising fees are prevented from sending bitcoin at all.
In theoretical terms, this is no big deal. In practical terms, however, it is an artificially added scarcity (certainly, bandwidth and processing power aren’t the limit there) that might allow your little Raspberry PI to participate but makes the network functionally useless as a payment processing channel.
I can’t say whether XT or the alternative proposals (like the Lightning proposal) better solve this problem, but to deny that the so-called “capacity cliff” is a problem seems to ignore the purpose of the blockchain as regards Bitcoin.
With something like BitTorrent, this restriction of using bandwidth wouldn’t matter — the time to download a file taking longer would be a minor inconvenience. With a payment processor that requires as close to instantaneous settling as possible and has to be able to easily scale (not just raise the price until the number of users drops off to some maximum), this is a problem of a completely different nature.