I've seen parsec (http://parsec.cs.princeton.edu/index.htm), which has a PDE pricing component, but the distribution is enormous and I haven't bothered to try to download it for review. I'm interested in finding benchmarks for pricing a portfolio of vanilla options and running some scenarios; has anybody seen anything like this?
On a single-option basis, there is this paper comparing methods by Mark Joshi. It doesn't specifically examine portfolios, but there's a reason for that. Portfolio and scenario computations are embarrassingly parallel, so once you have achieved your most efficient available option pricer, the rest is simply about wise distribution of your computational load.
There is also this somewhat minimalist paper for FPGA schemes, though to be honest I doubt many people are bothering with FPGA for option pricing, since you can be faster market-making on a delta grid.