Oke, let me rephrase. They used to use different methods but whether or not they claim standardised methods on paper doesn't mean it always works correctly. I know this for a fact because I see this fail often enough not to trust it.
Actually you were right the first time, I still see this happen today though admittedly not quite as often as it used to. When the standard was first written and implemented, too much latitude was written into the standard resulting in auto-negotiated mismatches. Lately I have seen this happen between Sun and Cisco equipment but have also seen this between HP and Northern Telecom equipment using crappy lan cable near the 100 meter length (max for the standard). Once again though, these are commercial installations and it does not happen often. For the average home users, it will be pretty rare. One caveat though, if I saw a network slowdown it is one of the first things I want to look at. That is also why I like managed switches because on a managed switch you can get to the switchport and see what has been negotiated and also see the lan statistics. Mismatched connections results in runts, giants, CRCC errors, etc. You will also see counts in these areas if you have a 10/half connection because after all, it was designed to operate that way, though your error counts will be quite low in comparison to a duplex mismatch where you have one side at half duplex and the other at full. Packets get through, but the connection is slow and there are many more errors.
Wireshark is not a network management tool it is a packet sniffer/analyzer and only displays packets that arrive at your NIC unless you are connected to a hub where all packets are distributed to all ports. In the modern world though we don't use hubs anymore because its an inefficient use of bandwidth and is less secure.
Lastly, if I were wiring a house or a new project I would use Cat 5E cable. We have found Cat-6 by any manufacturer has given variable results during gigaspeed testing. Cat-5E has been the most consistent.