If the encryptor offers multiple inputs, one a dummy "sure, here is my password, officer" HTML and passphrase, the other the sensitive one this scheme could work. Hiding both in a single blob, plus some noise, could work in the browser too. Seems it might offer some degree of plausible deniability. Some investigation of the decrypted dummy HTML might reveal it isn't large enough to account for the size of the blob. This doesn't seem like an insurmountable obstacle though. If the tool hides an arbitrary number of inputs in blobs that are indistinguishable you'd seem to have a great deal of deniability.

For those not familiar with the terminology, "rubber hose" in a cryptography context is a euphemism for torture. I know of two approaches to counter it.

Plausible deniability, where you can give out a password that unlocks some of the data, but your adversary has no way of knowing whether that's all of it. Useful if you can expect that you'll be let go if you're 'innocent'.

A self-destruct function, e.g. you give out the wrong code 3 times and your device wipes the data. Useful for military use cases where the data is worth more than the life of the person holding it.