If you combine fixed-point arithmetics, tiled memory access patterns, asm.js, and dynamic image processing pipeline compilation (the way that, for example, Virt

If you combine fixed-point arithmetics, tiled memory access patterns, asm.js, and dynamic image processing pipeline compilation (the way that, for example, VirtualDub did it), you'll probably discover that the "slow single-threaded language" probably isn't slow at all for image processing tasks. (And WebCL is still on the horizon.)