Hah, that is weird. Kind of a dick move, yeah.
On the upside, it's certainly no more than, somewhat less than 4.5V -- Rds(on) is given for this condition, and it's well saturated by then.
Offhand, I would guess typ +/- 1V, being a fairly common range. Logic level parts like this tend to have even narrower ranges (say 0.8-2.0V).
You can see from Transfer Characteristics, it's less than 2.5V (plots are typical data, by the way), mind, at higher Vds (10V) and current (graph cuts off at 10mA). Note the linear slope on the semilog plot -- this is in the exponential subthreshold* / cutoff region: we can extrapolate down to lower currents from here, with reasonable confidence. With a slope of about 10x per 0.35V, 0.25mA should be 0.56V lower, or Vgs(th)(typ) ~ 1.89V.
*Different meaning of "threshold". Vgs(th) is just measured at whatever low current. The device characteristic transitions from Id ~ exp(Vgs) at low currents, to Id ~ Vgs^2 at high; the transition between these characteristics being this meaning of "threshold". I suppose for small (RF?, IC) MOSFETs, such a current might be close to this characteristic, but they just keep using the same current for ever-larger power transistors, so, obviously they're going to fall lower on the curve, relatively speaking.
So, still no (max), other than what can be inferred from Rds(on), and similar shifting-about arguments using the other typical plots.
Tim