I don't think we should get complacent about strategies to both reduce the size of the Cryptonote blockchains and more efficiently swap old parts to disk without needing the full chain to be resident in memory.
Reason: The transaction volume of BTC is perhaps a little bit higher than that of XMR.
BTC: 50-70,000 tx/day https://blockchain.info/charts/n-transactions
XMR: under 2,000 tx/day http://monerochain.info/charts/transactions
It's not a problem when nobody's using the currency for anything except trading, but that's not the desired endpoint. The desired endpoint is more Tx than bitcoin -- and now is a good time to start engineering to handle that. Bitcoin itself is no paragon of transaction scalability as it stands, if you think the cool goal for cryptocurrencies would be to dethrone, e.g., Visa.
As to holding the block chain in memory... Lord no. This is just a bad decision on the part of the original reference or bytecoin code. I'm no expert here, but there are tons of robust scalable database solutions that run fine and fast from disk. SQLite for example would be a much better choice imo.
Visas database has to be gargantuan. Bitcoin and eventually cryptonote Will scale. Technology advances fast enough that the concern they will not is a bit chicken little. This is not a reason to avoid optimizing. And God knows cryptonotes database implementation needs an overhaul.