There's too many variables in the way, like the USB to X chip used in that converter plus the driver for it and how the OS scheduler handles the USB stack and API calls, all this causing major jitter in the communication timing.
In uni we still had air-gapped Win2K/98 computers for the robotics labs since in the days prior to Windows XP, your app could write directly to the serial/paralel ports, so you could have your program use 100% CPU load, jitter-free, to poll the serial/paralel port, that's why they were used in CNC. For example your Win/DOS program would directly drive the cutting head in real-time through the paralel port. You can say goodbye to that via USB converters unless you want your expensive machine mangle its cutting head.
My embedded systems prof from uni would only buy Fujitsu (Siemens) laptops as they were the only ones who still sold models with serial and parallel ports on modern hardware even as late as 2017(!), so if you're a professional or serious hobbyist in this space I suggest you grab one of those before they become scarce and overpriced like how high-end CRTs have now become.
There are companies in Taiwan/China making custom X86 barebones/motherboards using modern hardware and old ports but they're low volume and destined for the industrial market so they have eye-watering prices.
Many of those small industrial grade mainboards, especially the Mini-ITX ones, are also employed in game / slot machines and have plenty of serial ports, GPIOs, etc. There's a market for used ones on Ebay and elsewhere. There also are cheap PCI-E to RS232 adapters that cost less than 10 bucks delivered, so serial ports aren't really a problem unless one needs them on a laptop. Actually there are also moderately priced Mini-PCIE to RS232 adapters, but bringing flat cables out of a laptop doesn't seem that practical.
What is really annoying is how over the place the USB ones are in quality and drivers. The rub is they almost work. In many cases they are fine. But some of them have very odd modes of working fine then becoming jittery. What is also annoying is you can find a brand that 'works great' then the next batch they switched something out. Then there are some out there if your software does not play with the rts cts 'pins' correctly the whole thing will get out of wack. So your can end up in a condition where it works rock solid on real hardware but the USB will refuse to work correctly because you did not follow the 'standard', or worse sorta work.
For industrial use Siemens has a notebook called "Field PG". It has 2xEthernet and even the newest models have a real serial port (and also ProfiBus if you need that).
Check out how many ports and devices they can fit into a modern thin and light without compromising repairability
You can do 99.9% of "realtime" tasks with non-realtime systems that have been carefully tuned linux systems (multiple cores, remove all daemons, tune CPU scheduler) and I think thats a much better approach than trying to prop up old tech.
What I was referring to is that in the old days, PCs would control the CNC cutting head directly via paralel port, there was no microcontroller in the middle. These setups do not work wit a USB converters.
Quite often, CNC machining is not the first step in making a part. Thus scrapping a part is a very expensive, and could lose a client if it happens more than once after you "update" a machine.
There are very rational, very strong incentives to be conservative with computers used in industry.
I'm pretty sure they haven't updated the software on the elevator in my building in over 20 years but I'm fine with that.
They're not unreasonable in cost (start at like $100 for 2 ports), tons of features, and not flaky like the USB-rs232 adapters. Not so sure about latency, but throughput is fine.
Alternatively, you may take an embedded board standard like Comexpress that has a wide range of ports, including a bunch of real UART.
That doesn't stop their rampant use in those applications, regardless.