One of my next goals will be to do Ethernet to GPIB (thankfully most of the code will be reusable), but that will be after my OSHW function generator.
The GPIB part should be the hard part, bolting on any standard TCP/IP stack and ethernet MII/MAC should be fairly straightforward. Good luck with the project, most people undertaking it seem to spend much more time than expected getting it to work with various instruments. Gotta love standards.
It works with multiple instruments connected. When directly interfacing with my controller, you have to first specify the GPIB address of the target by sending "+a:1" without quotes, and where 1 is the address. My python classes automatically handle all of that.
Any particular reason why you chose to reinvent the wheel instead of just copying the Prologix protocol? Short of GPIB32/VISA support, this is the closest you get to a widely supported standard. There are even some open source language bindings available.
I include some basic examples with my python classes on how to use them. I do a basic *idn? query in one, transferring the 'scope waveform via binary and doing FFT (I've implemented binary block reading in the generic 'instrument' class, from which all others inherit from), and third transferring the currently displayed reading from my older Keithley 195. I'd probably do a more in-depth video tutorial to make sure everyone knows how exactly to use it.
A Keithley DMM wouldn't be my first choice for debugging a GPIB interface. Not that they're bad DMMs, but their GPIB implementation can be annoying. Fun like reporting an error several seconds and about five commands after the offending commands, or occasionally resetting and corrupting the output for one sample for no apparent reason when polled frequently. I tend to use HP equipment for debugging, since their implementation tends to be quite robust (being the inventors and all). Modern IEEE 488.2 or SCPI stuff is nicer than the old stuff when ROM was expensive and error handling was often limited to ignoring unknown commands.
GPIB can go much faster than 56k async serial. I've never measured it, but using GPIB was substantially faster than 38k4 RS-232 on the scopes I've used it on. I believe the theoretical limit is 1.8 MB/s for standard GPIB and even more for the high speed variant. Not sure how close real world instruments get to this. Prologix claim about 500kbps, but that's probably limited by the 8-bit AVR.