The deviation of the actual sampling rate from the nominal (8000, 11025,
etc.) have always been an issue for low bit-rate sound card modes. The
newer schemes may produce wider excursions, but that is a problem only
if you aren't correcting for it.
WOLF has an obvious way to compensate for the actual sound card sampling
rate, and once you have done it, the performance will be just as good as
it would be if the rate were exactly equal to the nominal. If you have
an accurate tone source, just use it in the WOLF calibration process,
and you'll be all set. Other tone sources include Loran-C and
divided-down outputs of higher frequency well-calibrated oscillators.
As you point out, the WSPR program does not seem to have a way to
compensate for far-off actual sampling rates. But WSPR's author, K1JT,
has a jack-of-all-trades program called WSJT that DOES allow for
calibration. WSPR is one of a number of modes offered in WSJT. So, that
might be a way out for those stuck with on-board audio setups.
That said, I've never had a sampling rate issue with WSPR on any of my
various computers, a couple of which sample pretty far off the mark.
If you think this stuff is fussy, try BPSK at 1 bit/sec where your only
adjustment is a trimmer cap in a crystal oscillator. Been there.