While I'm with you on the question of performance, I'd also question the suitability of FPGAs, both as an "open source" platform and as a learning tool for anything below university level courses. FPGAs are about as closed as it gets when it comes to hardware platforms. The verilog/VHDL compilers are, generally, closed source. I know there's an open one or two, but Mentor Graphics, Xilinx, and Altera all ship closed source compilers. The place and route algorithms that are used are all patented and closed source. The architecture of the FPGA itself is patented and closed source.
So, what, exactly, is the point of using an "open" processor on an FPGA? To make everything harder to do?
If you're really looking for a Free/Open processor, then your best bet is to put your money where your mouth is and back opencores.org in producing an ASIC version of the OpenRISC 1000. Even then, it's still built on a proprietary process in a fab, where you can't even get the technology files required to layout the processor without signing an NDA.
Here's the sad truth of it. You're dealing with a proprietary process anywhere from the chip level down. You simply cannot complain about not having open silicon and be taken seriously. Here's how it works:
If you want to make a chip, the first thing you have to do is find a design. Now you can make your own, and open source it, or you can get a pre-made design. If you choose to use an open-source design, then you're good--so far--but you'll have a significant performance lag behind the proprietary options. This goes double for video processing, memory controllers, buses, etc., etc.
Next, you need to find a fab who will make the chips for you. Here's where it gets bad. Even 180nm fabs consider their processes to be trade secrets, so that you have to sign an NDA just to get a process description file from the fab--this means your layout is, perforce, closed source.
Even if you somehow find a fab which will allow you to open the technology file, the placement and routing software for VLSI design is all closed source and patented. This is because place & route is a HARD problem. NP Hard, in fact.
So what it comes down to is this: until the homecmos people get their process going, you're stuck with something proprietary at some level. So then how much proprietary stuff is tolerable?
The Raspberry Pi Foundation had the goal of being bringing computing in a low cost package for education. The tradeoffs required to use open designs for the processor are quite steep: e.g. it would be a colossal time investment to get Linux running on a non-standard--read: non-proprietary--SoC. Using some proprietary chips to get there seems reasonable, so long as the OS doesn't become proprietary. The GPU blob is unfortunate, but not unexpected, particularly if you want decent performance.