I wonder if a remake of the Matchbox would be a good way forward - Jason, I think you offered to open up the designs, so it would need someone to be willing to do the buying and making, a bit of a thankless task... but if the net effect is bigger or faster, there might be a dozen takers for the feature set, as well as some takers who feel hardware beats emulation and who don't want to use a Pi. Also perhaps some completists, and some people who just want a nice FPGA dev platform.
Faster would mean bumping the FPGA speed grade from -2 to -3, for a modest bump in price, and/or swapping the RAM for a faster part, maybe 55ns or even 10ns, which doubles the RAM price. (Faster RAM might mean more stringent PCB design, which is another difficulty.)
Bigger would mean a different FPGA. Unfortunately, it looks like Xilinx don't do a bigger FPGA in quad flat pack, and a move to BGA is another difficulty in manufacture and of course a board redesign. As for size, the LX9 costs very roughly £13, the LX16 costs £18 and the LX25 costs £27. And those might even be plus VAT.
If an LX16 (60% bigger) is big enough for a 32016 with FPU and/or some cache, that might be a winner, but it's a new design and would be more expensive.
It's looking like faster might be feasible, but bigger is difficult.
Edit: looking at the various FPGA dev boards out there, Matchbox does stand out for having 32bit wide SRAM, and 5V interfacing, and reasonable cost (during its production runs).