I'm not really sure how to explain this without 'techno-babble' (or using my knowledge of materials science gleaned from professional training and experience), but it's a matter of how much steels are deformed, worked or stressed before fatigue cracking (see - can't do it without techno terms) occurs. Once it does in harder steels, the cracks tend to run fast, so failure happens fairly quickly. It's much slower in softer steels.
Spring temper is towards the softer end for heat-treated tool steels, so there is some scope for deformation without catastrophic failure. That's why setting a saw is OK. The limited angle through which teeth are bent also reduces the chances of failure significantly - setting teeth out at right angles and then reverse bending would be much more likely to result in failure, as would repeated oversetting and reverse bending.
Another factor that is often very significant is corrosion, and this does cause embrittlement. Any vintage saw with corrosion along the toothline (unless it's very superficial) will almost certainly result in broken teeth even if the attempted setting is quite minimal. The usual problem with such vintage saws is not oversetting but the reverse - no set at all! I agree with Deema here - corroded teeth are best jointed off and new teeth cut in good metal.
There's theory, and there's practice. I do know the theory (in outline - it can get VERY complex), but I've also had a go at the practice in this case. In a newish saw with no corrosion problem, overset teeth can safely be squeezed or hammered back without breakage. I've done it, as have many others, and it works. Just don't do it repeatedly on the same saw, and don't try it on an old saw with corrosion problems.
