If, OTOH, the government already specifically knows that you have incriminating data on an encrypted drive, this test doesn't seem to apply, per my reading. It's not "testimonial" for you to decrypt the drive, as they already know the evidence exists, and that it's on the encrypted drive. The plaintext doesn't give them anything they don't already know about.
How can one know something exists if they don't have it? They can be "pretty sure", but they can't "know". Therefore, providing the encryption passphrase is always testimonial. (Mumble, mumble, something about a radioactive cat...)
How can one know something exists if they don't have it?
I don't think the distinction the court is making here is particularly epistemological. The question isn't even directly about the specific evidence on the encrypted drive. It's about whether the act producing said evidence, itself, would be testimonial. If the prosecution "knows" you have this evidence, however — legally, of course — they came by that knowledge, then the act of producing it isn't testimonial. If they don't know of specific evidence, OTOH, then compelling you to produce any evidence you might have would be.
If, for example, you were dumb enough to admit to a third party that you keep the map showing where you buried the bodies on an encrypted drive, that person's testimony might be sufficient. Worse, you might have let slip that's where the data is while being interrogated. Or maybe the Customs agent saw a file named "XYZ Company Fraud.xls" the last time you came back from overseas, and now you're being prosecuted for defrauding XYZ Co. There are countless ways for the man to come by knowledge of the existence of a specific piece of evidence.
In the case of child porn, how about network logs from your ISP? Would this be enough? There's no way they can be sure that the images are stored on your disk. But digital traces are logged all over the place, so this is a very relevant point.
Look at the first comment by Jon Shields (on volokh.com, not here), addressing exactly that aspect of US v Fricosu (a prior recent case where the defendant was ordered to turn over a decrypted copy of the disk).
They had wiretaps of Fricosu admitting to someone else that specific information existed on his laptop. Although the prosecution did not have the plaintext documents that Fricosu was referring to, his admission over the phone was deemed enough for it to be a foregone conclusion that the documents existed on his laptop, and therefore the court could order Fricosu to decrypt.
Quoting footnote 27 of Fricosu:
[In the wiretap transcript], Friscosu essentially admitted every testimonial communication that may have been implicit in the production of the unencrypted contents.
The suspect could have told about it to somebody, or somebody (like informant) could have seen it sometime ago, or maybe even gave it to him while working undercover, or they could have observed the suspect receiving the file while working on the specific computer (i.e. via wiretap or by observing the connections on the sending side or while in transit). Of course, there's no proof that between that and current moment something didn't happen and the file wasn't deleted - but at least if it was not, the fact that it existed would not be news. That's like if I give somebody a secret document and he puts it in the safe, then the fact that he has the document in the safe is not news to me - even though in the meantime somebody could have broken into the safe and stolen it, for example, so I can't be 100% sure it's actually there.