(Seems like this should be r.r.a.homebrew instead of the antennas
group!)

Beware of the difference between a "power divider" and a "power
splitter." (The terminology different folk use can be confusing,
because there's not good consistency in the useage...) I gather,
without reading every word of this thread, that you want to make
network measurements, and the splitter is to provide levelling: either
a reference channel or actual amplitude control of the exciting signal.

So for a two-way, you want simply an input connector and two output
connectors, with a 50 ohm resistor from the input to each output. The
trick is that you want as near perfect symmetry between the two
channels as you can get. Ideally, each 50 ohm R will be in a coaxial
environment where the output end is 50 ohms and the input end is 100
ohms, but you don't have to control that very accurately to get it to
work well to 1.3GHz. With that arrangement, the source "sees" 50 ohms
if each output is loaded with 50 ohms, and if you use one output
channel to control the level, the level stays constant at the input
connector (since that's what's being monitored) and that is an
EFFECTIVE zero impedance point, much as the inverting input to an op
amp is a virtual ground. The the second output port then shows an
effective 50 ohms source impedance. You can show that monitoring the
reference channel instead of actually using it to control the level is
functionally the same for circuits which are not significantly level
dependent.

For a three-way, you want 50 ohms to each output port as with the
two-way, but now since the three outputs in parallel (each 50 ohms load
plus 50 ohms series resistor) result in 33.3 ohms at the junction where
they come together, you need 16.7 ohms in series from there to the
input port, to provide a 50 ohm load for the source (assuming 50 ohm
loads on each output). Note that 16.7=50/3. So you can make that
splitter with 6 50 ohm resistors.

Being just resistive devices, the only thing that limits the frequency
response is parasitic inductances and capacitances--not being able to
make the environment "perfect". But that's only a problem on the high
end, not the low end. DC is not a problem for these devices. However,
they are much more lossy than a power divider, which ideally has no
loss in the divider itself, at least with proper loads.

We had a need for some DC-6GHz splitters. The commercial ones we had
were rated to 3GHz, and indeed they weren't all that wonderful much
beyond that. A search for commercial ones that were rated to 6GHz came
up empty. So I designed a 6GHz one, using 50 ohm 0805 SMT resistors,
with some help from a mechanical engineer. The three output ports are
in one plane, radial spokes from the central node where all four
resistances come together. The input port is perpendicular to that at
the center. The resistors are all soldered together at one point, zero
lead length except for their terminations and the tiny ball of solder.
To get good performance to 6GHz, there's a screw that adjusts the
capacitance from ground to the central node. It's all housed in a
hexagonal aluminum block, with SMAs radiating out. I was able to build
a prototype that worked acceptably, using only PC mount SMAs soldered
together. I'd have a lot of confidence, based on that, that I could
make one that would work quite well to 1.3GHz even without much in the
way of test equipment to check it.

And you should have no trouble at all making some very decent loads
using the techniques others have mentioned. FWIW, I've had better luck
using two 100 ohm SMTs radially opposed than using four 200 ohm SMTs
with 90 degree spacing. I didn't investigate just why, but assumed it
had to do with the parasitics inherent in the parts.

Cheers,
Tom

(I'm a bit surprised you need a three-way splitter. Are you measuring
two DUT paths simultaneously??)