Why do I get a bad overall performance when using a Snmp instance for each Thread?

SNMP4J already uses multi-threading and non-blocking I/O operations internally. Using dedicated TransportMapping instances combined with a Snmp instance for each application thread does not improve overall performance for most applications. 

Instead use a single Snmp instance and share it across the threads. Using a single thread with asynchronous SNMP requests is also a valid and fast approach.