I still don't get it. The outside is dirty, right? He said in his post "You dip this probe into beer, sewage, or canned food a-stewing". So when you say "when really the window is just dirty" I don't get it - yes it will always be, because that's what it is placed in, no?
The permiability of the membrane would still be reduced by stuff on it, but as long as there is any permiability, the inner compartment will reach equilibrium eventually.
The big gain comes from a change in how you interpret the presence of electrons.
The older approach converted oxygen to electrical current, the magnitude of current flow relating to magnitude of oxygen depletion. The assumption built into that approach is that low oxygen depletion levels meant low oxygen levels, but that wasn't the only potential cause, because it ignored variation in the permiability of the membrane.
The newer approach equates current flow to oxygen concentration, as the system doesn't deplete the concentration any longer. The permiability of the membrane in this setup only contributes to a longer initial delay as the inner chamber comes to equilibrium with the surrounding concentration.
I think maybe one thing you have to consider is that sensors still require maintenance. Software can measure the length of time the sensor requires to reach equilibrium and send a maintenance required alert and someone cleans it (like if the software expects equilibrium in 10 seconds but the reading settles at 60 seconds, it can calculate the sensor is 80% clogged and requires cleaning). There's also all sorts of techniques that can be used to mitigate gunk depending on how the sensor is being used such as physical wipers, air-blast systems, ultrasonic cleaning systems, and chemical coatings. So as long as some oxygen can get in and an equilibrium is made between the fluid outside the sensor and inside the sensor, you'll get a reading that you can trust.
The sensor is normally placed in a dirty environment. This change prevented the need to calibrate the sensor for the level of dirt.
A dirty window only ruins the reading if you are measuring the speed of the oxygen passing through it. The three electrode design stopped measuring speed and started measuring balance. Unless the gunk is a total airtight seal (which is rare on the scale of an oxygen molecule), the sensor will eventually reach the right answer, whereas the old version would fail.