I don’t agree with such adjustments. The size of IP and TCP headers are not the only protocol overheads, TCP adds yet more slowness because it doesn’t drive the line to 100% usually. And which version of IP do you use, and what about the size of TCP headers, which is also variable? The user might not even be using TCP.
The figures quoted should IMHO exclude IP and TCP as they are just guesswork anyway, and the network providers should measure the true throughput in term of bytes sent in L3 PDUs that is where the cost of IP headers’ bytes are counted as part of the total bytes sent, and they should do this by fully saturating the line without using TCP at all. Different versions of TCP perform differently.
They can quote estimates for TCP throughput as well, but these should be a secondary item, needs to say exactly what it is "TCP payload in TCP/IPv6 with TCP timestamps" and they need ideally to have upper and lower boundaries for those figures, and point out that the TCP payload are very inaccurate figures that depend on the software you have at both ends.