The crystal filter can cause intermods.
One famous UK manufacturer selected filters for IMD: the best ones were for receivers and the less good ones went into exciters.
They also determined which way round gave the best performance.
It is usually considered that a crystal filter is a passive, linear, reciprocal, two port network.
Passive and two port - yes.
Reciprocal and linear - not necessarily, although it does depend on the performance you're looking for.
Mechanical filters manufactured before about 1975 or so were quite poor, too, but they improved after that.
Many years back, I measured a number of KVG XF9B 9 MHz SSB filters, and with signals at 10 and 20 kHz away from 9 MHz, the third order IP was around 15 to 20 dBm, depending on filter and whether the signals were 9010 and 9020 or 8980 and 8990.
Selecting filters for IMD is really not the way to go and is not the method used by me when I designed roofing-filters.
About KVG, we were the only importer in the USA for them during the 70s and 80s.
Why not ask the author what he meant?
Simple answer....He does not answer emails.
Signals farther away would be affected by the maximum losses of the skirt... controlled by the number of poles, not the filter bandwidth.
Actually, it is a function of both poles and bandwidth.
But too many poles and a narrower bandwidth cause havoc with the group-delay at the band-edges.
73 Jerry KM3K