ADDITIONAL_OVERHEAD_MAGIC_CONSTANT

I haven't got my head around that / got time to analyze that, but seems that the logic with the OVERHEAD_SIZE_BYTES (that the maximum payload is the MTU - OVERHEAD_SIZE_BYTES) seems to not be exactly right.

This magic constant for now helps, that there are no BufferUnderflowErrors (which happen at command/batch deserialization, if the payload has been cut off) occur.