In my humble opinion, without knowing the system it is indeed very difficult to know for sure. In general the factor of safety in-built in all the considerations while doing a theoretical analysis is enough to make the system work in actuality for long enough before it fails in real life, unless offcourse there are gross mistakes in design. If this kind of stress (99.9%) occurs in a sustained case we must attempt to reduce the stresses (play safe always). In an expansion case I might allow the results to pass depending upon the temperature and material. In occasional cases, I would hesitatingly pass. There also, it depends on the conditions of simulation and what is the type of occasional load.
Markl, I am sure must have encountered piping components which must have failed much earlier than governed by the rule he finally had deduced from their behaviour. Surprisingly, those results have been applied to a span much wider than his experiments.
So the best solution in a case of 99.9% may be to verify using FEA packages like FE-Pipe, etc.
_________________________
----