Looks like I answered my own question. It's well known that fixed-point IIR filters have the most difficulty when set for a low frequency (f << fs), and high Q. Thus I made a test of 2nd order single precision and double precision general filters as well as a state-variable filter, all set for low-pass, Q = 10 and f = 10 Hz. Read-back blocks show how each filter responds: The double-precision general filter is the only one that's close to accurate. Thus I conclude that the state-variable filter is single precision (or at least acts that way).
Bob