You are right that moving CF and serial port to Zero page can significantly reduce the size of program; move status bits to D7 and D6 also help. Scramble command and status of CF registers is probably too big of change (it actual is trivial to remap data lines, but I’m thinking of documentation nightmare).

My current approach is having the same design, but two different 32-byte bootstrap codes, one is CF boot and the other is serial boot. I can reprogram CPLD 100 times, according to data book, but probably many more. (Electron tunneling may damage the thin oxide causing reduced retention time, but for prototype, I don’t care.). I can serial bootstrap to load CF bootstrap and write to CF, then switch to CF bootstrap to test it. Once CF bootstrap is working, I’ll abandon serial bootstrap and use disk imaging tool to copy more CF disks.
Bill