But, any satellite based user will have a minimum of some 600 ms lost to latency, and which can increase during primetime...if satellite user's "proxy server" is located in same city as the test server...if not, latency increases as a function of distance from test server location.
Hence, unless a test site accounts for user's latency (not proxy server's latency), then results will be skewed much more on smaller sized test loads than on larger sized tests.
Hence, if a test site is using bytes/time formula without latency factored in, then inacurrate accounting.