Admittedly I am having trouble "unwrapping" what you wrote.
Yes, 15pF will be 100 ohms at 100MHz, but how many 100MHz sources of noise do you have that will induce even small amounts of common mode voltage? That common mode noise voltage will also need to present as a differential voltage at the receiver input where there is invariably a capacitor to shunt just such noise to ground. I can also use supplementary common mode filtering. Further, that 100Mhz noise would have to form a beat frequency with the underlying clock transitions to generate noise below the PLL cut-off. Not saying it can't happen, but it is what would have to happen.
A far worse issue with galvanic isolation is data induced jitter.
w.r.t. "RF holes", while that may be the case, it only needs to be the case at one end. You are also bringing in a very small piece of metal which only is an effective antenna at very high frequencies. On the transmitting end, there is no reason of course not to ground the BNC. On the receiving end, you can capacitively couple the shield to case with a small capacitance to eliminate RF, and get the side benefit of shunting the aforementioned common mode noise to ground.
Shields bonded at one end aren't good practice and a recipe to fail EMC. You can get away with a lot of crap for audio, I'm sure.
Yes, you can add some capacitance to the case but unless it's an annular feed-through type, it's going to be iffy. Amphenol does make BNCs like this but they are not commonly used. I've never seen a DAC use a common-mode choke on a coax input either.
We can debate the details, but optical is superior if you care about galvanic isolation. Since jitter does not practically matter as long as the PLL can lock and/or you have an ASRC after the receiver, I don't see much reason to use coax unless you need the distance or 192 kHz.