If you search DCRainmaker and/or GPLlama for smart bike power meter tests, you'll see they have done very solid comparisons of pretty much every smart bike out there against the power (often multiple) meters on their bikes. As you'd expect, some trainers are more accurate than others and in general wheel-off trainers (which cost more) are usually more accurate than wheel-on trainers. That's why some of the online racing rules require wheel-off trainers to be used.
It is one of those "you get what you pay for" and "you should pay for what you need to get" kind of things. Probably 90% of the riders on Zwift don't use power meters on their bikes in the real world, so any power numbers online are just for online purposes anyway.
I'm one of that 90%, been using Zwift for 4 years now but don't have a power meter on my road bike or my what is now called a gravel bike. I like doing races on Zwift as workouts, do structured workouts and lots of just plain exercise rides - and I like increasing my FTP on Zwift because it indicates my fitness is increasing. But the absolute value of my FTP on Zwift (based on my wheel-on Kickr SNAP trainer) is meaningless, since I'm not really competing in races.
When you look at all the others factors that aren't very accurate when you ride online (Zwift or any other), I'm not sure power meter accuracy really breaks into the top ten issues other than for pure online competitive racing!