Revisiting your original post are you talking about the difference between the number of messages reported by piaware as received from dump1090-fa and the number sent to FlightAware? (I use readsb, but I think the log entry is similar.)
May 16 16:41:38 pi43 piaware[942]: 6325827 msgs recv'd from readsb (2683 in last 5m); 6279878 msgs sent to FlightAware
If so, then the messages are not actually “lost” - there were a couple of recent threads on this, but I can only find one.
https://discussions.flightaware.com/t/piaware-only-forwarding-a-small-percentage-of-received-messages/87861
An older thread that explains what is happening is this one:
If I’ve assumed correctly, then there probably wasn’t a problem with running the R2 on the Raspi 4 (noting that you will get better results with a filter on either machine).
graphs1090 will let you easily see which position of filter and LNA gives the best results.
Looking at your airspy_adsb parameters, I think you should experiment with fixed values for -e and delete -C and -E. You are basically thrashing the processors - targeting 95% to be used by airspy_adsb but are very unlikely to be getting more than a fraction of a percent of extra positions than with much more conservative values. The -e value and other parameters I use are very conservative, but testing these settings on a batch of samples from my setup, showed I was missing only around 0.5% of positions and my Pi4 runs at 30% load. As I said previously also see what happens when you turn down the gain.
As @astrodeveloper says, there are no universally correct settings. It all comes down to experimentation - the rules of thumb for other SDRs often don’t translate to Airspys. That’s another reason to install graphs1090 and use the airspy charts to see if results get better or worse as you play with the parameters. (With the R2 you can also play with the 24M sample per sec rate - from memory people reported the Pi 4 struggled with -m 24, but your hardware probably won’t.)