Found this on a StackOverflow question from last year and simplified it.  Consider this makefile:

foo : recurse = +

foo bar : ; $(recurse)echo $@: $(recurse)

Here we use a target-specific variable to specify whether or not the "recurse" flag is set: this allows us to have a single rule definition which defines two different recipes which have different recurse flags set (something that's not possible any other way).

Now consider the (correct) behavior when we run each one individually:

$ rm -f bar foo; make -q foo; echo $?
echo foo: +
foo: +
0

$ rm -f bar foo; make -q bar; echo $?
1

For "foo", where the recurse flag is set via the target-specific variable, we run the command and it doesn't need to run any more commands, so the -q flag causes us to exit with 0.  For "bar", where the recurse flag is NOT set, we don't run the command and -q causes us to exit with 1.

Now if we run both together, we get this:

$ rm -f bar foo; make -q foo bar; echo $?
echo foo: +
foo: +
echo bar:
bar:
0

This is clearly not right: somehow the "recurse" flag set for the "foo" target has been transferred to / also set for the "bar" target.

I think that these targets are sharing too much information about the commands to be invoked and/or it's not getting cleared properly after each expansion of the recipe.