I think the relevant reliability would be the expected test-retest reliability of the measure at the two time points which are otherwise being used to examine difference scores. Thus, if you were looking at difference scores derived from an experiment where measures are taken one day apart where an intervention is applied on one of those days, then you would want to know the correlation between those two days in the absence of any intervention.
In general, I would question your assumption that hormone measures are more reliable than self-report personality measures. I'm not experienced with hormone reliability data, but a good personality scale should have test-retest reliability around .80 or 90. My intuition is that hormone measures have much lower reliability than this, but I'd be very interested in anyone who can cite test-retest reliability measures for various hormone measures.
I found this discussion of cortisol test-retest reliability (Nicolson):
Test-retest reliability of acute stress response measures appears to
be low, probably because of both the underestimated noise introduced
by spontaneous pulsatile activity (Young, Abelson, & Lightman, 2004)
and the tendency of the cortisol response to habituate following
repeated exposures. Low reliability of cortisol outcome measures means
that laboratory stress experiments are particularly vulnerable to Type
2 error. This issue is important in all studies, but especially needs
to be taken into account in intervention studies, where stress
reactivity is compared pre- and postintervention. More analyses are
needed to determine how many cortisol measures per session and how
many repeated sessions are necessary to obtain reliable measures of
stress reactivity for different subject populations and stressors
(Gunnar & Talge, 2007; Hruschka et al., 2005).