- JFETs are symmetrical devices*. There is no overlap between D and S directly, but they are separated by G. So it doesn't make much sense to specify D-S (or S-D).
*Well, surviving Si types are. And, maybe not completely symmetrical, but not dramatically so either. Like, 20 vs. 40V D-G and S-G or something. There are SiC JFETs available today, for power switching purposes, which are quite asymmetrical: 10 or 20V S-G, >600V D-G.
The same way we do for any other transistor. Note that BJT VBE isn't reliable; it's quite consistent for a given part, but varies modestly between parts, and most importantly, varies dramatically with temperature. In a single-transistor common-emitter amplifier for example, base voltage is fixed (by a voltage divider, say) and emitter voltage is allowed to vary, with emitter current fixed by a resistor. Thus the bias current varies by only the spread of VBE relative to the total emitter resistor voltage drop.
JFETs can be matched by buying them that way, or setting up in a circuit with a useful drain voltage (say 5-15V), fixed gate voltage (say 0V), and a constant current from the source (say 50uA) to a negative supply. The G-S voltage is then the pinch-off voltage (when the current equals the test current for that parameter).
If you were curious about why, the reason is that the channels formed between D-S, through the G layer, are extremely thin, formed by the diffusion of dopants into the silicon. While the doping impurities don't move very quickly during manufacture, the distance scale is very small, and the levels are very precise (local concentrations of parts per trillion make all the difference). So it's very difficult to control. Newer types, like BF862 (now obsolete) or CPH3910, as well as offering optimized performance (high gm, medium IDSS, and low capacitance), offer tighter parameter spread. (I don't know offhand if that's due to better process optimization, or alternative technology (epitaxy instead of diffused).)
High enough not to matter, i.e. avoiding the triode region.
Its practical significance is significant! The gate junction is a (normally reverse biased) diode, so can have quite low input current, whereas a BJT incurs full base current (Ic / hFE). This makes JFETs excellent choices for high impedance amplifiers, electrometers, voltage buffers, etc.
It's noteworthy that IGSS generally increases at high VDS, due to hot-carrier effects I think (basically, the current flowing through the pinched (saturated) channel, can flow so "fast" that some electrons (or holes) tunnel out of the channel and fall into the gate). For least gate current, this voltage should be kept modest.
By extension, JFETs can also be used as low-leakage diodes. G and D+S are oppositely doped silicon, so if you tie D+S together, you simply get a diode from G to them. Not a good one, mind -- the internal resistance is comparable to the channel resistance, and maximum current is limited (IG(max) might be 10mA). But within that range, the leakage current can be much smaller than say 1N4148.
