I've been working on speed and stability issues.
I'm pleased to announce that I have a rock-solid protocol in place now, which is at least 7 times faster than the initial version. Much time spent with a logic analyser and a verilog reference have paid off massively.
I can load highway encounter - a 40k program - in just over a second. Overall it takes longer to load the transient program from floppy than it does to transfer the code from SD card.
I have a new prototype board finished which is working nicely. This will be sent to a beta tester for some real-life work and I'll report on the results here later.