The Thymio horizontal IR sensors use some hardware and software filtering to remove the external noise, this is a feature :)
There is first an optical filter tuned on a specific wavelength: 940nm. Thus if you want to interact with the sensor, you must use this wavelength.
Then the sensors have an hardware electronic high-pass filter with a tau ~= 15us, thus by pointing an IR emitting a constant light, you won't "see" it on the Thymio, as it will be filtered by the electronic before reaching the microcontroller.
And then, there is some software processing which make sure we read the sensor value only when the sensor is actively sending light. This is needed in order to not have several Thymio interfering with each others when they are close ( less than 1 meter ) to each others.
So what you experience is perfectly normal, by pointing an IR led (or the sun or an old-fashion lightbulb BTW) to the Thymio you should not impact it's sensors values.
The last behaviour that the sensor see "nothing" when the IR led is switched-on really close to it can be explained by the sensor saturation. All the above explanation hold only if the sensor is not saturated.
When too much IR light hit the sensor it will saturate, which mean that the sensor value will not change even if the intensity of light is changing. Since we have a electronic high-pass filter, we will simply see as if there is no light hitting the sensor.
