This is the first chance I've had to look into this further. I went and checked my set of original FBD filters for my Duplex. I purchased my unit new before they were discontinued so I know they were OEM pieces. They showed no indication of Roscoe product or serial numbers. Could it be that the construction of the filters changed over the product lifetime of these safelights? Mine was purchased quite a few years ago.
I then had a look at the spectral transmission graphs made available online by Roscoe. There are two very slightly different versions of the #19 Fire filter listed. I've referenced the version I purchased. Here are the three in question for others who may not have seen them:
Rosco E-Colour+ #019 Fire
Rosco Roscolux #3406 Sun85
Roscoe Roscolux #3407 Sun CTO
And here again is that LPS line spectrum from Brian Niece, Associate Professor of Chemistry at Assumption College in Massachusetts (fourth item down the page):
Low Pressure Sodium Line Spectrum
From the generated line spectrum of an LPS lamp the extra blues and greens caused by the inclusion of the argon/neon Penning mixture are readily apparent. And a look at the #19 Fire filter chart shows that it should be almost perfectly opaque to those wavelengths all by itself. For standard OC-rated b&w materials this filter should work fine. And in practice, at least for me, it does.
On the other hand, I'm not sure what additional safeness the other two filters would add, given their respective charts. Especially in the green regions for VC papers. Filter #3407 in particular allows a substantial percentage of blue-greens and greens to pass. This should be an obvious risk factor for VC papers if the #19 were not present.
Might these second two filters be intended for different materials? I'm really curious regarding your observed results.