Thought I would chime in here. I have used both the UMC 1820 and the Scarlett 18i20. You can achieve better round trip latency with the Scarlett depending on your system. On my desktop I use a pci express card for USB audio devices, all the USB processing is handled on the card and there's no sharing of bandwith with other usb devices. This will go a
long way in helping to get low buffer settings to work smoothly and dropout free. With the Scarlett I was able to get the buffer to 32 and it was stable for me (but see below). The lowest the UMC 1820 will go is 64 but I really couldn't hear much of a difference when monitoring between the 2 at those settings. Both units sounded very similar, the Scarlett pres are a tad bit warmer to my ears, but I use several external preamps so that's a not an issue. The Scarlett is bigger and built like a tank, but the Behringer is not a cheap build at all the only real drawback being wall wart power supply. Both units handle ADAT without any issues. One thing the Behringer has that I like is each of the 8 channels have line/inst switches as well as pad switches. The Scarlett only has those options on the first 2 channels. The Behringer also has a knob to adjust between the direct signal and the input monitor signal from SONAR whereas the Scarlett only allows control of that in their software. Not a big deal either way, but the software with the Scarlett let's you set up a lot of nice mixing options for monitoring. The Behringer doesn't come close on that.
Unfortunately both me and one the other producer where I work had to return the Scarlett units. We both had issues with the devices losing connectivity in our DAWS and both of us had issues with odd noises coming from the unit when engaging the transport from time to time. He was using his on a MAC with Pro Tools, I was using it with SONAR on a PC. Not sure if these were just bad units or there were driver issues. So, while I still think the Scarlett is the better unit, the Behringers have performed better for us and the cost saved us some $$. Your results may (and probably will) vary.