Well, I can't actually give you 'real world with ePMP' on our network, since we're still deploying and don'thave any large client count sectors yet. However, I can tell you that we've used Canopy 900Mhz stuff for years, and with only 3-4 Mbit on the AP for everyone to share, it'll do a remarkable job of scaling everyone and keeping latency predicable. I can also tell you that when I went to the ePMP Training Session in Edmonton last fall, there were a couple WISP's there that had changed out from Ubiquiti to ePMP and they had 60+ SM's on each Access Point, and they were thrilled.
Of course, they were 5Ghz (not 2.4) and they were doing relatively low plas of 3 and 5 Mbit plans only, and their install criteria insisted on every install being -60 or better, and if not they left the low signal guys on thier other technology and so on. But they felt that 60 was quite doable as a client count as long as the install criteria was strict. But wiith other gear, 20 or 25 is starting to stretch things.
And yes, latency can potentially certainly increase depend on a whole bunch of things. If you understand how the scheduler works, the Access Point basically looks at everyone who needs to receive or transmit data, and the AP give them a time slot for their data. So, it's basically saying ''SM#13, you can transmit in 320ns, SM#17 you can transmit in 480ns" and so on. So, if you have 40 clients and they all need to transmit data at the same time, and if you have them all set on a plan with 10 Mbit download and 2 Mbit upload, and if you've got your ratios set to 75%/25% and if you have 5ms frame size..... well then there is only certain math that works - there is only 25% of that 5ms frame that is the upload slot, and at some point there is no math that makes it all work out. There isn't going to be 40x10 Mbit download and 40x2 Mbit uplaod all a the same time on that AP - so something is going to give - that's just physics.
But, what the Cambium ePMP scheduler does is an 'air fairness' so that everyone get's a fairly fair split of the available air time, and it'll try to slow everyone down fairly and not let someone close and loud monopolize the air time, and it does it in a way to minimize ontention or colissions or ACK's (which waste air time for everyone on other gear). The AP and the SM's know when they should listen and when they should talk, and they don't need to do a 'clear channel sense' to see if the channel is clear or not, and then maybe transmit and maybe still bump into another customer who's transmitting at the same time and so on.. The Cambium ePMP scheduler does a much much more organized job of all this.
Also, with almost all other makes/brands, every packet that is transmitted is 'ACKed' from the far side, so as distance from the AP to the SM's start to increase, performance decreases - and it decreases for everyone, even the close guys. So it's CPE#1 talks and waits for it's ACK. CPE#7 talks and waits forit's ACK and so on. While each CPE is waiting for it's ACK's, nothing else can happen for any other CPE either. With Cambium ePMP, the AP knows how far away the farthest SM is, and when the AP transmits it's data to each SM, it also tells them when to send back their ACK's - so the Cambium Access Point basically does SM#1 here is your data (and wait X ns), SM#7 here is your data (and wait Y ns), SM#24 here is your data (and wait Z ns) and then all in sequence, all the SM's can then send back an ACK in the 'ACK schedule'. Much much much more efficient.
So, people are often talking about 802.11AC an dhow much faster that could be - and that's partly true. However, there are tons of 802.11AC gear out there that wastes 1/3 of the potential air time with collisions, senses, ACKs and so on. 802.11n gear that wisely uses 100% of the potential of the chipset can end up being much faster in real world, much lower latency in real world, much more scalable in real world.
Oh... I'm talking a lot.... I think I had too much Coffee today... ;)