Electronic – Antenna matching network. What? Why? How

Without more information about your design I can't comment on the matching techniques. But here is some help for your test setup.

You can attempt to make relative measurements, but RSSI is a poor parameter for verification because you have to make sure you're in the linear range of the RSSI for the comparisons to be somewhat similar. Be aware that RSSI is just a current measurement of how saturated the amplifiers are and in general is not very accurate.

A spectrum analyzer would tell you much more and be able to give you a quick idea of how your device performs across the band.

Barring all that...you'll need to test this in an open area, outside, to avoid multipath interference. The more open and lacking of objects (especially metal), the better. Make two movable test stands (plastic or fiberglass is best) with batteries for power. Measure the distance between your reference receiver and your test transmitter (DUT).

Start with the manufacturers reference and characterize it at low/mid/high bands. Make sure your RSSI readings are stable and your background measurements (with no transmitter in operation) are quiet. The ISM900MHz band can be very noisey, so be careful of false readings using RSSI. You might find you're constantly turning on and off the transmitter to check background levels.

The RSSI needs to be in about the middle of it's range. If it is not, move the units closer or further apart. Measure this distance carefully so you can repeat it in the future...and don't loose your reference receiver! Also keep track of the orientations used and make sure the cables are carefully laid out, taped down for repeatability (both receiver and DUT). You might even want to lay down some copper screen material under the receiver and DUT to provide a more consistent ground plane for testing. Try to run the wires down through the center of the screen under the device.

Then take your design at the same distance and repeat the measurements across the band. With RSSI +/-3 dB is probably about all the accuracy you could expect. And since your design is physically different you might need to test a variety of orientations to insure you're getting consistent results and not testing at some antenna null or lobe.

All else equal, the differential traces will radiate less. This is because their opposing electric fields cancel for distances "far" away, with "far" being relative to the spacing of the traces. Less radiation also means less opportunity to pick up or contribute noise to or from other components near the feedline, which may or may not be a concern for you.

Given this, it could make more sense to use differential lines for as much of the distance as possible.

However, to get the advantages of a differential feedline requires that there are no common-mode currents on the differential traces. If there are common-mode currents, then the differential microstrips are effectively just a fatter ordinary microstrip. Ostensibly the IC should drive the differential traces in such a way that there is no significant common-mode current, but you could check the datasheet to see if they have any suggestions. Differential traces are harder to route, so you have some design trade-off decisions to make.

The reason microstrip radiates more is that the trace isn't fully enclosed, and so the electric field extends away some distance from the trace. Here's an image from High Speed Signal Propagation: Advanced Black Magic

This image also hints at another solution: stripline. Co-planar waveguides are another option which can be fabricated on a 2-layer board. Each of these helps to constrain the electric field to a smaller area, reducing radiation loss and increasing isolation. They are probably overkill for your application, but for instances where you need a single-ended transmission line and microstrip just isn't good enough, they are options.