Unlike fission power plants there isn’t going to be years worth of the stuff on site. For one thing 5% of the stuff decays sitting around so you want a tight loop of production to consumption. Further, the reactor is holding low density plasma so there is very little inside at any one time.
That's not how it works. A low molecular weight gas anywhere below the homopause remains well-mixed, and does not separate by molecular weight. On Earth, that's anywhere below 100 km altitude.
As a beta emitter it’s blocked by just a few feet of atmosphere thus rendering the bulk of it harmless. Eventually, some will combine with oxygen and end up as water, but again most of that just ends up in the ocean.
Once in the troposphere, hydrogen of any kind will be oxidized to water within a couple of years, and then rain out. Of more concern would be accident processes that would cause it to be oxidized immediately. For example, any fire or exposure of hot materials to air would cause associated tritium to react.
This is all unrealistic anyway, since the plant will not have 100 kg of tritium on hand at any time. That's about the amount consumed in a year, but the reactor could not afford to have any substantial amount sitting around, decaying, or else the breeding ratio will be too low.
Anyway, for a more detailed description of what I was expecting. A least initially it’s going to act just like hot air. Hydrogen’s is 7% the density of air so Tritium is presumably 21% the density of air. Which is similar to air at 2000f without the particulate matter of smoke.
About 100kg should be roughly 500 cubic meters depending on temperature, but a more reasonable limit of ~10kg is still close to 50 cubic meters of gas. If we are talking a sudden release from say a pressurized tank rupturing outside that’s going to from an invisible but mushroom shaped blob and rise. Where a detonations mushroom cloud stops rising as the temperature cools, this thing only slows as it mixes with air which isn’t that fast. The troposphere is only ~8miles up so it’s likely to reach the stratosphere mostly intact.
If we’re talking a venting pipe or something that releases gas more slowly then you get much faster mixing. However, baring the slowest of leaks we are still likely talking going up hundreds of feet at a minimum and more likely miles before it dispersed enough to act like the rest of the atmosphere.
In the absolute worst case, you still get a lot of vertical mixing of the atmosphere from thermals and rapid dispersion from the wind which doesn’t slow down. Within days you’re talking thousands of cubic miles of atmosphere. So, a short term evacuation of those down wind might happen, but they should be able to return in days.
A 1 GW(e) reactor would burn enough tritium in a year that, if that quantity were to be released into the environment, it could contaminate 2 months worth of the flow of the Mississippi River above the legal limit for drinking water.
Tritium containment at a fusion reactor will have to be damned near perfect. This will be a major problem, as tritium permeates through all sorts of things (for example, plastic seals on containment penetrations cannot be used).