I almost entirely agree, the difference is minimal (especially if a support agent can easily recover it, that's a pretty low bar). Encrypting passwords and keeping the key in an offline (or hardware-protected), highly secure location helps manage your risk but "don't ever do it" should still be standard response when considering a system which lets you recover passwords.
I can think of two situations where design dictates you need passwords to be recoverable, but even though they exist you can (and I do :)) argue the design is wrong:
- Online aggregators like Mint/Yodlee/etc. that pass through your credentials and use screen scraping to get your data. Obviously a faulty design but in the financial aggregator case it's kind of your only choice and many customers are willing to accept the tradeoff.
- I've heard policy enforcement (e.g. "Your current password must not contain any of your N previous passwords and must meet [set of strength criteria]") used as an excuse but I don't really buy it. Not having access to the original password limits your choices for future policy enforcement (to what you could derive from the password when first set) but with a little foresight and extra work you can do fine just with hashes.