Popper, your maths appear to make many assumptions.
For a start, who says that all your applications are using an acknowledged protocol (TCP)? Some could be using an unacknowledged one (UDP) where no return path is needed.
Also, for TCP the maximum packet size is 1500 bytes whilst the minimum is 40 - a ratio of 37.5 to 1. This is more than the 33.3 to 1 of the mooted service but TCP has an ace up its sleeve. To improve bandwidth, it supports a scheme called 'windowing'. This allows the sender to transmit several packets to the receiver before it has to wait for an acknowledgment. At this point, provided it has received them all correctly, the receiver can then send one ack to cover the lot. So the upload speed does not have to be faster than 1/37.5 of the download speed after all. That is not to say that all implementations will be this efficient!
In a previous existence (before traveling through a brief wormhole) I was an architect for an implementation of the OSI link and transport layers.
http://en.wikipedia.org/wiki/Transmi...ntrol_Protocol