Okay, interesting results, and I don't know how meaningful they'll be to you:
When the OPUS file is loaded, it has a sample rate of zero, and an automatic loopstart of 0 and looplength of... the track's length in seconds, as a double (that is, number with a decimal place). This happens to be the length of the original OGG file in samples divided by its sample rate. This may be salvageable, but it means you'd have to give the loop properties in terms of timestamps instead of sample numbers, which may be less precise.
So Starboard Authority has a length of 7,631,360 samples and sample rate of 44100 Hz. Its loop data is set to loopstart: 265,970 and looplength: 6,881,514.
The OPUS file gets a natural "looplength" of 173.04672916666667, which is its (very precise) length in seconds. So what happens if I skip over the sample rate math and set the loopstart to 265970/44100 = 6.031065759637188 and 6881514/44100 = 156.0434013605442?
...
Failed to set the 'loopEnd' property on 'AudioBufferSourceNode': The provided double value is non-finite.
Well, that's disappointing. What the heck is "loopEnd"?
EDIT: Oh. That's stupid. It tried to add the two numbers... as strings, which of course doesn't work. Why the pfargtl did it do that?!