What you seeing is very typical. Much depends on the process technology and normal process variations. Most suppliers will spec parts with max Vdd = 3.6V, but do always build some guardbanding into the process, so that in reality parts can operate at higher voltages, eg 5V.
Parts will generally fail because of EOS, electrical over stress. This results in damage to the very thin gate oxides rupturing (punch through) and then transistor fails.
What about the parts that do not fail? Well, the engineer must carry the risk. Much depends on how long they were exposed to the higher voltage. They will have been stressed beyond the manufacture's recommended operating conditions, and there could be a real risk that parts may fail somewhere down the line.
Analogy is much like hunting an animal. Best is to shoot it dead. The wounded animal may continue walking for days before it drops dead. Or it may recover and live for a long time.