cap4000
your statistic is far from clear. Is it that 1 in 500 to 1 in 1000 beams ultimately fail under the LRFD design regime, or is it that 1 in 500 to 1 in 1000 failed beams under the LRFD design regime fail due to typical loading conditions.
I am not mistaking an incipient failure for statistical probability. One of my friends has, conveniently enough, been researching probabilistic engineering. What he does is compare the actual capacities of the beams with the design capacity of the beams. These capacities are represented by a normally distributed bell curve. When the capacity curve is lower than the design capacity curve, a failure happens, this happens to occur in a very very small part of the curve, typically outside of 5 standard deviations, any less that say 3 standard deviations, 99.73%, and the code would consistently design members that failed. I am not suggesting that we should design to capacity, which may well be the 3 or 4 SUVs that I mentioned above, the structure in all likelihood has broken a serviceability limit state at this point and despite the fact that the bridge can hold 3 or 4 SUVs, the SUVs cannot possibly drive over it. However, this bridge that we designed for a single compact car can in fact hold up 3 or 4 SUVs and therefore it is acceptable to allow 5% wiggle room beyond the code.
However, I don't need statistical evidence to show you that the code is conservative enough to allow for 5% beyond design capacity. Consider LRFD.
1st the design values for determining the available strength in the first chapter of AISC 13 you may notice that there are always two values, this is because the values that the engineer is supposed to design with are the smallest recorded values for each member. The other values are the values that the product has been specced for and they are the values that the detailer should use in design.
2nd consider that A36 doesn't yield at 36 ksi like the code says it does it is expected to yield at 39.6 ksi instead. In this step alone you gain ten percent strength for all things A36, I'm not sure what the expected yield factor, Ry, is for A992 but I'd believe that it is somewhere in the 1.05 to 1.15 range.
3rd you magnify your already conservative estimates for your loading. Although I do agree with this step, mostly because I can't see into the future to tell what the actual loading will be.
4th you aren't allowed to take the full strength of your member, it is reduced discriminatorily based on what failure mode governs the design. Although I agree with this part too, mostly because it gives credence to the age old term "run time".
Ultimately its pretty clear that your members are designed much stronger than they need to be. I am therefore comfortable saying that 5% beyond design capacity is ok.
