Hi.
Yes, on this Access Point, I am the only client. This was our first test AP when we were first evaluating ePMP vs some other brands. To describe this setup, this is right in downtown, the AP is on the top of an 18 story appartment building, and shooting right across residential town to my house about 4 KM away. I previously had a different brand of 2.4Ghz link to my house, and it would struggle to get 1Mbit sometimes. At other times I could get 3mbit or 5mbit if I waited till 3 AM to use it - but it was suffering from interference pretty badly. I can hear a bazillion routers in 2.4Ghz in the site survey, so I wasn't too surprised.
So, I fgured it would be a good test for ePMP - I honestly figured it would fail about as badly, but figured then we'd at least know the worst possible case. So, I switched out the AP to a ePMP Access Point (same mount, same sector, same everything) and I switched out my SM from the other brand to a ePMP (same mount, same dual pole grid, same everything) so that this was JUST a comparison of Brand X to Brand Cambium.
My performance went from ~1 or 5 Mbit, to about 100 Mbit throuhput. I get ~70 Mbit down and ~30 Mbit up when testing to our local SpeedTest server at the shop.
http://community.cambiumnetworks.com/t5/ePMP-1000/EPMP-UBNT-Change-Over/m-p/48913#M4744
http://community.cambiumnetworks.com/t5/ePMP-1000/The-ePMP1000-latency/m-p/48671#M4662
As for your other questions, I see very little real-world throughput difference from one to the other as it varies a bit from test to test and day to day anyway. However, 5ms is probably 10% faster overall I'd guess, and it just seems to work better for me. I think the 2.5ms was primarly designed to allow better syncronization with the older Motrorolla PMP100 radios, and no so much for lowering latency. Your mileage may vary though...
If lowest latency is the most important thing, then you can choose to NOT syncronize and you can lower the latency the most, but then you defeat much of the magic that is ePMP. If you want it to be scalable and if you want to try to mitigate as much self-interference as possible, then you're going to want to add in that extra few ms of latency and you're going to want to GPS Sync everything with Fixed U/D Ratios.
For me, I'm normally on 5ms frame size. For me, an extra 7 ms of latency plus or minus is nothing - we are probably 50 ms from Google.com through the internet anyway, so to me it doesn't much matter if that's 50 + 8 ms, or if it's 50 +15 ms - there's not much difference.
I'm running FIrmware 2.6 and I haven't really had any issues - it just works and works and works, despite the cold, despite the interference.