I hope no one has already posted this somewhere in the past but this thread is so long that's it's almost impossible to determine if it's been mentioned somewhere. So here goes....
For several years I've been using two stacked 91XGs, combining them with the lowest loss 2-way splitter I could find. I wanted to improve upon this situation as any loss in combining the antennas subtracts from the ~2.5 dB gain you get by stacking. It's easy to lose a dB or more.
My 91XGs are the older ones with the ferrite baluns. I have measured the newer ones with the PCB balun but it was no better, about .5 to 1 dB loss depending on frequency.
Recently I was introduced to the idea of the broadband coax balun and I realized it could also apply to the 1/4 wave matching sections. I had ignored these devices before for combining UHF antennas because I didn't think they were broadband enough. I was wrong about that.
I constructed and measured 2 UHF 300 to 75 ohm coax baluns and a 1/4 wave UHF 2 way power combiner and I measured their losses. These are the losses I measured on the bench using an HP-8640B signal generator and a Boonton 9200 power meter.
Coax balun - 0.5 dB @ 470 MHz, 0.3 dB @ 585 MHz, 0.4 dB @ 698 MHz
Power divider - 0.12 dB @ 470 MHz, 0.25 dB @ 585 MHz, 0.4 dB at 698 MHz
I retrofitted the 91XG plastic feed boxes to accept the baluns and replaced the 2 way ferrite combiner with the coax combiner. I measured between 1.5 and 2 dB improvement on channel 28 using my spectrum analyzer. Channel 28 is the only LOS station I have near me that I have monitored for a long time to be sure it is stable under varying conditions. Most of my stations are 2 edges and vary a lot with conditions so I cannot use them for accurate measurements. My overall impression is that the numerous other stations are as strong or stronger than I've ever seen them. It'll take some time to get a good feeing for those stations under varying conditions.
For construction I used Belden 8241 (RG-59 with a copper shield), Belden 9269 (93 ohm copper shield) and RG-223 (50 ohm double shielded cable). I couldn't use RG-6 because you can't solder to the shield. The 93 ohm cable is used to make the 1/2 wave balun. It's as close to the ideal 150 ohms that you can get these days. I used the RG-223 as the 1/4 wave matching section. 50 ohms is very close to the geometric mean of 75 and 37.5 ohms. I found the Belden cables on Ebay and I had RG-223.
I've attached a few images. The first one is of the 2-way power combiner. I modified an old satellite DC power inserter to join the 3 coax cables with very short leads inside. I used coax seal to keep out the water. The RG-223 length is 3.3" and attaches directly to the preamp input.
The second image shows a complete balun minus the ferrite beads. The 1/2 wave section is 8.5". It is attached to the 91XG with the leads as short as possible.
The third image shows the balun in place on the 91XG with two ferrite beads acting as a UHF choke. They are Amidon FB-61-6873 beads.
The preamp I'm using is a Tinlee MA-25U-77, 30 dB gain and 2 dB noise figure. They tell me it's typically 1.5 dB.
I think I've squeezed everything I can out of these antennas.