Concerning QoS performance of the different devices, my experience (and testing) is as follows:
ER-X: 180+Mbps with Smart-Queue (with approximately 500 flows - using torrents to simulate multiple users simultaneous connections).
ER-L: 70Mbps with Smart-Queue (with even only one flow).
With the ER-X and 500+ flows maxing out the connection, the internet itself is still very usable - even for VOIP, but some pages and images that have large files can load more slowly, as one might expect with a fully taxed connection.
You certainly should have better QoS performance with the ER-X than with the ER-L, in my opinion, when QoS is concerned, the ER-L is not an upgrade.
I do not understand why you would max out at 40Mbps on a 100Mbps pipe with QoS on the ER-X. Granted, and this is important to take into account, the value entered in the configuration for smart-queue bandwidth is automatically multiplied by .95 before being applied - that is to say your actual maximum bandwidth would be 95Mbps if you set it to 100Mbps. [I'm fairly certain it's .95 - it could be .90 though, perhaps
Now, I could see if your 100Mbps connection were being utilized by multiple computers that you may only get 40Mbps on a speed test, but the aggregate usage should be relatively close to the maximum.
So, I said all that to say I think there may be something else going on here. This in particular concerns me:
very small office, just a few users - with SmartQueue turned on the 1.8.5 firmware ERLite, the WAN Download speed capped at approx 10-20Megs. When I turned off the SmartQueue, it popped immediately to saturate and give me the expected 100 meg download throughput.
I'd want to do a test at this site when no one else is using the connection - to verify if simply turning on the Smart-Queue lowers the bandwidth that drastically (with only one connection it absolutely should not). I'd look at the number of connections with 'sudo conntrack -C' and the number of flows through the device with a 'tc class/qdisc show' command. I'll update this later when I'm home and able to form the exact command you can do here to give you insight. I've been told that we may see some statistics from Smart and Advanced Queue built into the EdgeOS command system at some point (but I'm not holding my breath).
I'd also look at 'top' to see the CPU usage (system interrupt: si) of the device while performing tests - if si isn't high (<40%), your device is not being taxed and should most certainly be capable of higher bandwidth with QoS enabled.
It may be helpful for you to post the entire configuration you've been using to test with both the ER-X and ER-L, as well as a condensed version of the results of your tests. Something like:
Site 1 (100Mbps down / 10Mbps up fiber with 10 devices):
Directly connected to modem / ONT - 99Mbps down, 10Mbps up
ER-X - xxMbps down, xxMbps up
ER-L - xxMbps down, xxMbps up
One thing to consider at your larger site would be to increase the number of flows in Smart-Queue. Something like 2048 or 4096. This could help with making sure multiple different connections from different computers aren't grouped into one flow. Also definitely check the AlliedTelesis unit (I'd even get them to send me a new one if it wouldn't cost anything).