@Siwastaja: Assuming R5 is not removed, err . . .shouldn't you be using the Thevenin equivalent resistance for the time constant calculation?
Obviously yes, but then again the time calculated would indicate the time to reach 63.2% * 6.0V = 3.792V. With R5 removed, source resistance is now 100k and not 50k, but the voltage after one time constant has passed is now 63.2% * 12.0V = 7.584V.
In both cases, the ESD diodes are already clamping the voltage before one time constant.
R5 will help dissipate the capacitor faster.
If reproducing the exact duty cycle, or delay, is of any importance, I would recommend simulating the circuit in Spice. MCU input pin can be modeled as two generic schottky diodes, one going to GND and other to ideal +3.3V voltage source.
If you want to keep R5, then yes it's a good idea to lower its value to form a better matched voltage divider. Just don't go too far not to eat into logic level margins. Use the lowest possible actual value of that "12V" and make sure the output is then at least 3.0V or so.