Actually combining blockchain pruning (i.e, the "main" blockchain only storing UNSPEND addresses), and some of the new emerging technologies like ed25519 that even on an old 2010 CPU we're already capable of 80,000 verifications per second. Seeing how we have a HUGE HUGE way to go before we actually reaching that amount of transactions per second, when we actually do reach it would likely be way post 2020, where we will have even faster hardware and likely better algorithms, which should technically be able to sustain the guidelines I mentioned.
Not to mention that most people would be running light clients anyway around that time, just as described in satoshi's original paper.
In the end there is no reason not to have everything running on the main blockchain since technological difficulties can be overcome.
I've heard all this kind of things ad-nauseam for the 3+ years that I've been following things. Simultaneously I've seen both the hardware and network improvements which have been made realistically available over that time period, and also the struggles Bitcoin has had just to make the original 7 TPS number kinda work. Many of them were not pretty. Some of these rosy predictions about how technology will swoop in to save the day (and let us go from 7 TPS to 200,000 TPS which is enthusiastic even for your type) would be more convincing if there were a few more signs of it actually occurring.