Ok. And can we see what the estimated parameter values are (i.e., mp) in the 2 cases? That might help explain the behavior you are seeing.
I am definitely no GP expert (that would be @bwengals ), but I might recommend avoiding pm.find_MAP() and instead using trace = pm.sample(). find_MAP() is not recommended in other (non GP) settings and is used, for example, in this GP notebook.