This is vaguely correct behaviour.  The tar format uses a record of
all 0s as a terminator, so when tar(1) sees it, it cleans up and
exits.  If the tar file was originally written with a larger block
size, you may not read all of the 0s at the end before tar(1) exits,
which would leave gzip(1) with a broken pipe to write the rest of them
to.