It’s mostly a numbers game - dump1090-mutability picks up a lot more candidate frames, most of which are discarded because they don’t pass validation.
Given enough candidates, some errors are going to slip past error detection.
The message error correction coverage in the Mode S message format isn’t great either - it gets worse as you have more aircraft with bitwise-similar ICAO codes - which can cause odd effects (squawk/altitude getting attributed to the wrong aircraft is the most common case; undetected errors are another possible result)
Any idea what VRS consider a malformed frame? The message framing should always be OK.