That's unrealistic for most modern stacks. Miami needs more than 512k with a GUI. Genesis looks like it can do with closer to 256k. The 68000 does not have as good of code density as the 68020+. A useful TCP/IP program is needed on top of that like a browser. Miami will run with 2.x if you choose an older more basic GUI. Genesis probably needs 3.x for Reaction or MUI. A 68000 with <1MB and KS1.3 is barbaric. Please upgrade if you want to run anything besides games.
A lot of things can be done with 68000 + 1 MB. But you have to skip the fancy GUI stuff and the upgrade hysteria. Using different binaries for 68000 and 68020 is also common and thus solves the code density problem.
A useful TCP/IP program may be Telnet and FTP, and definitely distributed filesystem(s).
Point is being able to use the same software foundation for all Amiga models. That reduces the number of source code modules to keep track of.
Genesis works with Reaction or MUI. The AmigaOS 3.9 version uses Reaction as default. Although AmigaOS 3.9 requires a 68020+, it looks like Genesis is compiled for the 68000 and can use the 68000 AmiTCP kernel. There is a lot of compiler fluff in the main Genesis program. I have a working modified version for the 68020+ that saves about 15k in program size and I haven't even optimized in the 68020+ integer multiplication and division yet. It would probably be possible to save 40k as the optimization is poor. What bugs do you know about?
MUI seems overkill for basic connectivity. And the bugs I read about on eabime (or something alike it).
I did find some very suspicious code in a function starting at $148b0. It looks like this:
lab_148b0:
link a5,#0
move.l d2,-(sp)
move.l (8,a5),d2
jsr lab_232
move.l d2,-(sp)
jsr lab_23a
nop
dc.w 0,$4e55 ;this would be an ori.b #$55,d0 but it's obviously not
dc.w 0 ;this would also be the start of a ori.b with the next function used as the data
lab_148cc:
link a5,#0
move.l d2,-(sp)
...
This function would likely result in a crash (illegal instruction, illegal address or similar) or at least trashed memory as it falls into the next function with the wrong offset. It's strange how such code could be produced by a compiler. I created a fix in my modified Genesis program although I haven't figured out what causes this function to be executed yet. It is called (jsr) from 4 places in the program. Any other known bugs?
Personally I prefer to have the source code and fix things there rather than to rely on magic binary changes to fix things.
A basic network package would be configured using textfiles. Provide connectivity using driver layer, MAC to IP mapping via ARP, IP routing, and UDP/TCP. Then DHCP and WLAN configuration can be available as applications. Such that Telnet/FTP works. Anything else would be optional.