The buffered output 66 byte buffer is in addition to the default TX 63 byte buffer on UNO to give a total ~130 (less a few bytes set aside for ~~ buffer full marks)
Due to some 'fancy' coding, writing to buffered out writes through first to the hardware TX buffer until that fills up, and only then buffers the rest in buffered output
You can use
output.getSize()
to get the combined size of buffered output and hardware TX buffering