One thing the FPGAArcade has going for it is far faster memory, so even if the AGA implementation has no enhancements, the 256 colour modes should perform a lot faster simply because there will be far less contention on the memory bus.
The blitter should be significantly faster (unless they slow it down for compatibility) and would work with the faster gfx memory to give even a bigger speedup. The RTG chunky modes will reduce the amount of work needed also providing a speedup.
Wonder if you can put faster chip RAM in an A4000 to help the performance in AGA?
I believe the Boxer was going to have faster chip ram. C= was brain dead for not constantly upgrading the speed of the custom chips and chip memory. The custom chips were an accelerator on the 68000 but a bottleneck for some operation on an Amiga that had a 68020+. AGA did upgrade the speed over OCS/ECS but not enough for how late it came.