In practice, this type of approach is only useable for very close range applications. First of all, different types of ``white'' paint may not reflect the same amount of IR! If a maze is painted with different types of white paint, calibration becomes impossible.
In addition, micromice are suppposed to work in a normal environment with ambient lighting. If hot lamps are used, they also produce IR. How can a robot distinguish IR from background lighting versus its own IrEDs? Well, you can use modulation to screen out some of the ambient light. However, this does get somewhat ugly in terms of signal to noise ratio.