...It depends on height and tuning of the stub. The same stub at a different radiator length at the same height would create differing radiation efficiencies between the two antennas. At a lower height the shorter antenna may work better, since it is closer to its ideal height for its design...
I don't think there is as much variation in practice as you might think. Certainly, while
the feedpoint impedance will vary somewhat with height above ground, the resulting change
in SWR will make little difference in the antenna efficiency, or the efficiency of the feedline.
There may be an "ideal height for its design" in terms of getting a perfect match, but the
greatest variation with height will be in ground losses, which will be roughly the same as
a function of height for most types of horizontal antennas. Really the only optimization for
height is a function of vertical angle of radiation, which has nothing to do with antenna
efficiency, and also applies equally to most horizontal antennas. So the 204' double G5RV
will have the same angle of radiation broadside as the 102' version - there is no significant
difference between them in terms of "optimum height for its design."
There are lots of other things that also can affect the SWR for the same antenna raised up
a few feet: the way the matching stub is dressed, proximity between the ends and buildings
or other wires, the angle in the midpoint, etc., and these will vary from one band to the
next. So the SWR may go up on one band and down on another.
Probably the most significant difference between the 204' and 102' versions will be the
azimuth radiation patterns: this will be slight on 80m, but much more on the higher
bands where there will be a complex pattern of lobes and nulls, such that which antenna
seems better for any particular contact will depend on the distance, direction, frequency,
and ionospheric conditions.
The behavior of the G5RV is pretty predictable in terms of impedance and radiation pattern.
You can get a little variation due to how it is installed, especially if the wires are sloped
downwards too much. The choice of wire and stub lengths are designed to be a compromise
that gives a usably low SWR on 80, 40, 20 and 12m: while you can often load it with a
tuner on other bands, the coax losses are likely to be high, as shown here:http://www.vk1od.net/antenna/G5RV/index.htm
Another major source of quirky behavior is the lack of a balun at the junction between the
matching stub and the coax: common mode currents on the coax can mess up the pattern
and, if the coax is laying on the ground, increases the losses. (This isn't the fault of the
antenna itself, but how it is fed.) While Lou Varney G5RV himself originally recommended
against using a balun, that was due to the poor performance of the baluns available in
the 1950s. A good 1 : 1 current balun at that point will often make a big difference.
Because the tuning is a compromise, you'll likely need an antenna tuner on most bands,
regardless of what some manufacturers may claim. Their insistence on a minimum of
60+ feet of coax will improve the SWR by increasing losses. (True, G5RV recommended
this, but it was because the tube-type transceivers of the time had a limited ability to
match impedances below 50 ohms: his recommended coax length transformed the low
impedance up to a workable value on 80m. This isn't a limitation for modern wide-range