The subprocess module only uses the file object to get a file handle by calling the "fileno" method. See Issue 19992 about documenting this. For Python to compress the output of the child process, you would need a pipe.

Gzip file objects provide the "fileno" method, but it just returns the underlying file descriptor. Data written to that file descriptor would normally already be compressed by Python and goes straight to the OS. There is also Issue 24358 opened about whether "fileno" should be implemented.