it is not exposed to more than 3V3 and no ringing. Why would the I2C being exposed to a connector be a potential issue in general?
By "traditional zener type TVS" do you mean to use an actual Zener instead of a TVS?
I2C on connector is bad for EMI and ESD. Fine if well shielded (as the HDMI example (Display Data Channel (DDC)), also used in VGA and DVI -- all use shielded connectors and cables), but the low bandwidth may give the false impression that it doesn't matter. But there's hardly anything special about I2C transmitters or receivers; there's a minor amount of filtering (maybe 50ns worth) and that's about it. Extra transitions corrupt the state easily, pins don't have extra robustness towards ESD, and EMI banging into ESD clamp diodes shifts the baseline which can also corrupt logic levels, even if the RF isn't being read directly as signal.
Besides, it's no good for long cable runs, not that that's relevant here I guess, but cable length is an easy way to soak up extra loading capacitance compromising clock rate.
Finally, it's not easy to filter, because the impedance isn't constant. You want to minimize loading capacitance, but you can't get the cutoff frequency very low just by increasing inductance: when a driver pulls down, it'll ring (because R(driver) < filter Zo), and then when driver(s) turn off, it'll rise slowly again -- the edges are asymmetrical, which I mean, you already know this, but trying to put a filter on it just doesn't work very well for the same reason all over again.
It truly is well suited to exactly what it is -- onboard communication!
There are extenders and translators out there, if you absolutely need to use it over longer distances (e.g. differential I2C, which physically works something like CAN, but it's the same old I2C protocol)... or there are other standards to choose from.
![Smiley :)](https://www.eevblog.com/forum/Smileys/default/smiley.gif)
As for TVS, the regular kind is just a zener, a beefy one that's rated for surges. You can indeed use regular zeners, if appropriately rated; they might not be, so it's simply safer to get the TVS version. They might well put the same chips into both products, give or take appropriate testing; who knows.
There's also the snapback diode (the unidirectional one). This has a negative resistance characteristic, and, it's not clear if it'll turn off with the pullups in there; they don't specify the valley (turnoff) current or voltage.
You'll also see low-capacitance TVSs, which are of the usual (zener) type, but have a series diode so that the TVS capacitance charges up to signal peaks, only loading the signal initially. Often the TVS comes out on a pin that can be tied to VDD, so it stays out of the way normally, and also protects VDD. See dual and quad arrays for USB and such.
The other kind of negative resistance TVS is the SIDAC type, which has low leakage until breakdown, then stays latched on until holding current is released. Advantage of course being much higher surge energy handling for a given voltage rating. They aren't available at such low voltages though (I think?).
Tim