Another advantage of this stateless client concept is that it seems that it would be easier for anyone to be a validator or full node. This fits nicely with Ethereum’s ethos of decentralization. Compare that with a proposal that the number of validators in Casper be restricted to 250, or whatever. With a stateless client perhaps that would not be necessary, and perhaps anyone could be a validator. However, I haven’t considered in detail the practicalities of anyone being a validator, or not having a limit on the number of validators, with this concept. Reading after that full nodes store modified state trie objects are stored for three months, then forgotten randomly by clients after that, this would then place some minimum limit on the storage space required by full nodes, thus it wouldn’t be so easy for anyone to be a full node. “We expect the number of “archival nodes” that simply store everything forever to continue to be high enough to serve the network until the total state size exceeds ~1-10 terabytes”. 1 or 2 TB can be done economically by anyone, but not all desktop computers have 1 or 2 TB of space, particularly computers that only have SSDs like mine (although you could use an external hard drive, but that would reduce bandwidth via a USB3.0 connection), 10 TB is harder.
“Clients that wish to ensure availability of specific pieces of data much longer can do so with payments in state channels.” This idea is not a regular payment like rent, but it does internalize at least to some extent the cost of storage. However, more accurately internalizing the cost seems like it is worth further investigation.