Thanks Rust -- good point. Though I'm still not sure that in our real world application of recording PA systems it matters, though there clearly is a difference in theory. My understanding is that the point you make only would figure in if the noise floor of our taping system is more than the noise level of the PA system we're trying to record. (I guess we could talk about recording the whole room, not just the PA stacks, but the noise of the people in the room, the HVAC system, etc would probably still lead to the same conclusion.) Or perhaps stated differently, if the PA system is only providing 68db of dynamic range, it doesn't matter if we limit our recording system to 84db instead of 96db by pushing our levels.
Also, as you note, I don't believe there is even a theoretical difference in dynamic range in recording at -3dbFS peaks (or even -6db peaks) rather than 0dbFS or 0.5dbFS -- you still use that last bit. And even less of a difference if you record at 24 bits and dither down to 16 bits.
So anyway, I agree with you on what you said, but I'm still worried that tapers who don't understand this all that well will start pushing the hell out of their levels and getting clipping (thus doing more to reduce the dynamic range). All to get better dynamic range out of their recorders in theory, even though that given the noise level of the system they are recording -- representing the absolute level of dynamic range that is achievable -- is well within the limitation of their recorders whether at -12db or -6db, let alone -3db. Bottom line, I still think that a few db's of headroom will make for better recordings, not worse.