You left off the one that i would persue - multiple UARTs with a single PIC as the traffic cop to and from SRAM. I believe you can come up with a design that does not need to centralize the data in one processor.
On the higher spec approach, you might want to take a look at the Renesas line of micros. the 16 series has several versions that are designed to hook directly to SRAM and have built-in DMA controllers. Though, they may have enough internal RAM to handle your application and thus not need external RAM. iirc, there is at least one with 4 usarts built in. Good speed, 16 bit operations and Decent C support.
how much assembler programming have done to date? If you've got decent experience in asm on any machine, then it won't be a stretch but if you have done very little, I'd stay away from it as you are significantly increasing your overall learning curve. I know other will disagree but you know yourself best.