Windows ReadyBoost allows removable drives to be used as swap space to improve system performance. The typical usage model seems to focus on using higher-speed USB flash drives for this purpose.
As much as I hate to let Microsoft decide anything for me, I long ago learned that (as long as I'm running a Windows box) the OS will generally be much happier if I let it configure it's own automated settings for swap allocation.
The USB bus can be fairly crowded, and even top-end chipsets don't approach the theoretical USB2 HighSpeed 480Mbps limit (for example, Intel X58 only supports up to 180Mbps total bidirectional USB bandwidth, probably less in real mobo implementations). Still, real performance can be faster than most SATA and many SSD drives. I don't know what impact SuperSpeed USB3 will have, since I've never actually used it. Yes, I've seen SuperSpeed USB Flash drives, but like their HighSpeed predecessors they advertise that spec to announce compatibility with the USB standard, not actual benchmarked or theoretical performance.
My idea is to use a HighSpeed USB controller and banks of RAM instead of NVFlash ... the logic being that RAM read/write access is generally on the order of < ~5-10ns, while flash is > ~35-60ns read and > ~100-150ns write. Assuming the drive-controller doesn't bottleneck things too badly, a "USB RAM Drive" should theoretically be much faster than a USB Flash Drive. It also doesn't have limited write-cycles before failure. It's even a great way for the paranoid to permanently destroy their cache and temp files.
Implementing this sort of thing isn't much of a problem for me. I've never worked with the Arduino parts popular on this board, but I'm confident parts like an ARM7 or even a Parallax Propeller (which I can use) will easily exceed the capabilities of the CPLD controllers used in consumer flash drives. Sure, RAM costs more than Flash, but in this instance I just happen to have piles of old SDRAM and DDR1/DDR2 (along with spare mobo sockets...) which aren't worth anything anyways.
Of course this wouldn't be a casual project, it would take time and effort. Before I invest that effort, I'd like to hear feedback for this idea.
Is it worth the effort? Will I see any real performance gains over using a standard fixed/removable drive?