Seems to be a SNR issue. See the 3rd paragraph.
Passive devices typically, but not always, have a low input impedance and high output impedance compared to active devices.
Any high output impedance has more difficulty driving long interconnects so the shortest distance between that and the load is preferred.
Why an attenuator is suggested at the amp is more complex than I'm willing to type, other than to say swimming downstream is easier than upstream.
I have, and use, the Rothwell's to reduce the Cal Alpha's 3.5v output to match other sources. In this application I find it effective. The only down side is that it seems to slightly roll the high frequencies a tad, not necessarily a bad thing for the Alpha. I tried it between a pre-amp and amp and, as suggested by Rothwell, it did lower the pre-amps noise floor and give a better range of the volume control. Now the 'why', at least my speculation.....
The output of line stage of the pre-amp was 26v - the Alpha 3.5v. The effect of the insertion of the Rothwell (-10db), while constant, on the higher gain unit it was relatively 'less' and more difficult to detect.
It is also possible that a high level source might overdrive a pre-amp input and the use of an attenuator would reduce resultant distortion.
FWIW there are pre-amps with input gain reduction circuits (my SP 10 has a 6db switch on the front panel) which accomplish the same thing as an attenuator when used between the source and pre-amp. It provides a better range of volume control but doesn't reduce the noise floor.
I probably would not use an attenuator between the amp and pre-amp only because its an inellegant solution, not that its seriously flawed in any way.
Hope that helps a bit.