Thanks @tauzero - towards the end of chapter 16 there is a comment that to achieve an overall Butterworth response in a multi-section filter, each section should have a different Q. That implies that the copy-and-paste sections in the FDeck and Elliott designs are approximations, and the WinISD calculated responses are correct.
I'll have a play in Win ISD and see if I can get a two section filter to produce the same response as their composite one.
@nekomatic - as far as I can tell, the Q defines the sharpness of the corner, and is not related to the slope of the curve further downhill, so for example with a high Q, low order filter, you could have a high peak at the corner followed by a relatively gentle slope.
David