My kids used all three IR sensors for line following at the local competition. They worked, but delivery was so slow that by semifinals they edited the code to just drive straight for the tower without them (bump switch to end). That method was much more efficient, even if they missed the tower and had to restart.
For Regionals, they changed the sensor layout and their location on the robot. Delivery improved some, but it was still more efficient to just drive at the tower. We kept both options available in the code, but the drivers never once use line following during the competition.
We talked about editing the code to ignore a stray sesnsor error (one that oscilated too quickly) but never implemented it. Maybe I'll task one of the coders to work on that in the off season.
I was watching one highlight video from a local competiton and the lines disappered from the courses halfway through the day. It turns out that no one at the competition (large enough to have 2 fields) was using line following and the tape was peeling and causing problems. They removed the tape and played on.
FYI, we 3D printed a housing for the 3 sensors that kept them separated to reduce interference and isolate reflections. We probably need to experiment with it to get more than just anecdotal results on its effectiveness.