The proxy will send each resolver their respective key ranges, if either resolver detects a conflict then the transaction is not committed.
This has the flaw that if only one of the two resolvers detects a conflict, the other resolver will still think the transaction succeeded and may failure future transactions based on its write conflict ranges.
In practice, a well designed workload will only have a very small percentage of conflicts, so this amplification will not effect performance.
Because of this effect we generally only recommend increasing the number of resolvers if you are running into a performance bottleneck because of the resolver.