Actually, some of what has been brought up about curve guardrails here suggests a different function than the straight track bridge/trestle guardrails. For those straight guardrails, from what I understand, the backs of the flanges never touch the guardrail for the purpose of preventing a derailment. Instead, the guardrails are just there to guide a derailed wheel in a wheelset as it runs on the ties in order to prevent the pileup that often follows a derailment.
I was not aware of a guardrail being set up to actually contact the back of the flange in order to prevent a derailment. That would be a function identical to switch guardrails, which will contact the backs of the wheel flanges if they are tracking far from the stock rail, and pull them back toward the stock rail in order to keep the opposite wheel flange from picking the switch frog and derailing. But I have never seen a continuous guardrail on a curve set up that way. It would have to be positioned with its head only a couple inches from the head of the running rail. I have never seen a photograph or actual case of such a guardrail, and have never heard of it before this thread.
What I have seen on curves are guardrails set far enough from the running rail to allow the wheel to fit into the gap. So those would be guardrails to guide derailed wheels in order to prevent pileups, but not to prevent a derailment in the first place.
I have seen guardrails on curves placed inside the inner running rail of the curve in some cases, and outside of the outer rail of the curve in other cases. As someone explained earlier in this thread, I think the reasoning for placing the guardrail on the outside of the running rail in a curve is to prevent ice and snow from plugging up the flange space, and to allow the flanger to work through the area without needing to be raised to clear the guardrail.
The disadvantage I see in placing the guardrail outside of the running rail is that it is nearer to the end of the ties where the most checking occurs, and there is the least holding power for spikes. Spiking near the tie ends also encourages checking and splits.
Since derailments on curves are encouraged by the centrifugal force, oftentimes only one guardrail is used because the centrifugal force only goes in one direction. If the guardrail is between the rails, it is outside of the inner rail of the curve. If it is outside of the rails, it is outside of the outer rail of the curve.
In the photo posted above by Todd Hackett on 12/9, there are double guardrails, apparently only to give added assurance against a running off and piling up on the outside of the curve. Being that both guardrails are outside of their adjacent running rail, neither of them can prevent a pileup by cars running off to the inside of the curve. So one of the two guardrails is a backup in case the other one fails to catch a derailed wheel.
And both of those guardrails are too far from their adjacent running rails to prevent a derailment. They can only guide the wheels to prevent a pileup after a derailment has occurred.