Of course, instead of just theorizing about which factors have the greatest
impact after changing the client's MTU, someone could end this discussion by
performing the following experiment with:
* PC with MS Windows
* Macintosh
* Unix box
Hook each box up to the same dial-up line, using the same dial-up service.
Measure response time to visit various web sites, ftp sites, etc. Perform
various runs with different MTUs.
Of course, to ensure validity of the data, memory and disk caches must be
cleared between tests, and the tests must be run multiple times to remove the
effects of caching at the servers and between the servers and the client.
If the PC shows a significant performance improvement at smaller MTUs, but
the other two don't, we can conclude that Windows is unusual (in a new way).
But if they all behave similarly according to MTU size, Windows would be
"innocent."
I would be very interested in the results. Unfortunately, I don't have the
equipment at home to do this. Does anyone out there?