Fundamentally, a MOSFET isn't a switch. A switch is strictly on and off (within various meanings of "on and off", depending on what kind of switch you really mean..). A MOSFET is a linear device with varying levels of "on-ness".
A more useful definition of a MOSFET is, a variable constant current sink, with a backwards diode, and which looks like a variable resistor when the voltage is low. The current is controlled by Vgs.
A reminder that there is no absolute voltage. Vgs is always gate to source voltage, and Vds is always drain to source voltage. It doesn't matter if you stick a resistor under the source, or any other kind of network or source, because gate and drain are still relative to source, period. If you measure gate-to-ground voltage instead, and source-to-ground voltage is nonzero, you've failed reading comprehension, because you measured something no one told you to measure...
What the datasheet tells you is, for this-much drain current, you need Vgs of so-and-so. Vgs(th) is measured at the given Id. The current is small, usually 1mA or less. Think of this as the point where it becomes sensitive; much less, and current just stays more and more zero. At Vgs = -20V, you can't get any more zero than zero* (and you'll be measurably there already by 0 or -1V, as it happens). But going up in voltage, it becomes more and more sensitive, and draws more and more current (or exhibits a lower and lower resistance, to a point).
*Not actually zero, but leakage current. Usually in the nA to pA. They rate it at 1uA or something like that, but only because it takes time to precisely measure very small currents. Almost no one is going to care if a power-switching transistor leaks 1uA, anyway.
The intent of the Vgs(th) figure is, this is the offset where it becomes sensitive. Your circuit needs to accommodate this variation, in order to use currents around this level; or for switching purposes, you need to deliver enough excess beyond this point to ensure it looks like a "switch".
Consider a faucet. It goes from off to on over a variable range. The amount of pulling or turning needed to go from the fully-seated or fully-off position to a dribble is the threshold. It's different for every faucet, as I'm sure you've noticed (and some are more irritating than others). Likewise, it's different for every transistor, and varies with temperature by the way. (Transistors don't happen to exhibit the even more annoying characteristic that many faucets do, where it goes, say, off -- dribble -- full cold -- still cold -- a little OHGODSUDDENLY HOT -- still hot -- more hot -- end stop.)
Low voltage MOSFETs (<= 20V) are more sensitive, so can be reasonably rated for "switch" duty with fairly low voltages, like 2.5 or 3.3V logic levels. For these, Rds(on) at Vgs(on) is just the same as any other device, but with smaller numbers.
Anything 30-600V that says "logic level" will reach its rating at 4.5V, nothing less. (And anything for power or speed will do much better with >8V anyway, so "logic level" should never be used as an excuse for poor drive in a power application.) Due to Vgs(th) variation, the amount of Rds(on) (or Id if pulled into the linear range) at, say, 4.0 or 3.5V, is not easily defined, because it's not apparent how close the transistor is to Vgs(th) just by looking at it.
Tim