Good point, for there is (AFAICS) no spot in the standard that makes it possible for a UCS's second SCS not to be the identity conversion if it invokes a constructor. In fact, the only place that explicitly specifies that second SCS is [over.ics.list] (p6, p7). Maybe the sentence you quoted was composed by simply enumerating all kinds of functions invoked by user-defined conversions? Perhaps the author of the section wanted to avoid any (more or less) profound exclusions?
Btw., it appears the wording in your quote is weird anyway, because a constructor is a "user-defined conversion function" (at least as implied by [over.ics.user]/4). Although, then again, that term isn't normatively defined (and "user-defined" could be taken as a separate adjective)...