member and findall or setof don't get along?

The following fails on the first recursion at the findall:
cd1([Disk | Ds], SoFar, Names, WithDates) :-
	format("Ancestor Graph ~p: ", Disk),
	deep_ancestors(Disk, now, Graph),
	format(atom(FileName), "ancestors-~p.svg", [Disk]),
	u_plot(Graph, FileName),
	pgraph(Graph, WithDates), nl, nl,
	vertices(Graph, Vs),
	%setof(Name, W^member(nw(Name, W), Vs), Names),
	findall(Name, member(Name, Vs), Names),
	append(SoFar, Names, SoFar2),
	cd1(Ds, SoFar2, Names, WithDates).
It also failed with the setof.  So on the first call it works; then it
recurses and fails.  In the debugger it simply fails immediately at the
line, even if I try to step in.  But see below for a variant in which it
succeeds only if stepped into with the debugger.

When I try to reproduce with a small example, using the value of Vs
showing in the debugger just before the failure, everything works:
weird(Names) :-
	Vs = [ nw(littleboy,now),
	       nw(pdb,now),
	       nw(pdb5,now),
	       nw(dpath([lv(backup,littleboy)]),now),
	       nw(path([oscorn1, usr, local, var, spool, bacula]),now),
	       nw(lv(backup,littleboy),now)
	     ],
	setof(Name, W^member(nw(Name, W), Vs), Names).

Also, using
	setof(Name, cd1a(Vs, Name), Names),

with the helper
cd1a(Vs, Name) :-
	member(nw(Name, _), Vs).

fails at the same spot if I step over ('s') setof.  But if I trace into
it (' '), the entire operation succeeds.  Then the next u_plot fails.

commenting out u_plot and pgraph did not help.  Vs is not empty and
consists entirely of nw(_,_) items according to the debugger.

Running 6.4.1.