This has been discussed all over the place since CF and SD cards have dropped in price.
If you move the Windows XP pagefile to a drive other than the hard drive that the system is running on, you gain some performance. Why? Supposedly, it's because the drive heads don't have to be accessing the pagefile at the same time they're looking for the program file.
Windows Vista does something like a mirror of the pagefile in their ReadyBoost technique, using a USB card reader to cache system data. The problem there is that the USB interface is pretty slow compared to an IDE or SATA bus, so it's not the best solution.
Windows XP does not allow one to mount a flash card as an internal drive through a USB card reader, at least, not without some serious gyrations, and you're still stuck with the relatively-slow USB bus.
So how to do it?
I have been working with two variations over the last couple of weeks, both camping out on the Secondary IDE bus, to avoid any data bandwidth conflicts with the system hard drive on the Primary IDE bus. You gain nothing moving the pagefile to a different partition on the same hard drive, and if you want the best I/O speed, you really should move that pagefile to a different IDE/SATA channel.
One way to accomplish this is to use a Compact Flash IDE adapter, which is a small circuit card that mounts a Compact Flash card on one end, with 40/44 pin IDE connections on the other end. There are jumpers for Master/Slave, and a separate power feed. This is the one I'm using:
Windows XP recognizes it as an internal hard drive, but there's a catch - it has to be
Industrial or
True-IDE flash. If not, then Windows XP will only recognize it as a removable flash drive, and not allow one to move the pagefile to it.
Once installed, it was easy to simply relocate the Windows XP pagefile to the flash "drive", remove the original pagefile from the system drive, and pretty much let it go with no further ado. The pagefile for this particular system is approximately 2Gb in size, so there's another 2Gb left on the card. I've moved some files in and out, and even loaded some videos to see what the performance was like. Writing was about like any other spinning platter hard drive, but reading files was pretty darned quick. I'll try to get some hard numbers later using PCPitstop.com, just to see what the transfer rates are compared to the rest of hard drives in this system. I was NOT able to format the flash drive in NTFS, so I left it as FAT32. That's fine, because NTFS does a lot more reading and writing in file management, and reading a pagefile every now and again via FAT32 isn't really a problem. I've removed all the files save for the pagefile and Windows-associated System Volume Information folder, so that the Compact Flash card's wear management features have plenty of available transistors to shuffle amongst.
How well does it work? Keep in mind this is a multi-cpu Xeon 3.0Ghz machine with 2Gb of memory, and I doubt that it ever really hit the pagefile that hard to begin with. However, I have noticed so far that Windows boots a bit faster, and applications load a wee bit quicker. I watch the green access LED on the IDE adapter, and it doesn't really show much activity save for boot-up and starting larger programs like Quark Express and Adobe Photoshop CS2. Brothers in Arms Earned in Blood seems to hit it a fair amount, also. It's actually working pretty well, and I've since installed it in my wife's identical workstation, with the intent of doing a long-term evaluation to see just how well the solid-state CF card holds up. So far, so good, and it's really a minimalist install, although I mounted the CF/IDE adapter in her IBM's hard drive bay so I could see the three LEDs flashing through the front grill.
On Mtnbkr's suggestion, I also looked into a SSD hard drive, also known as DOM, or Disk-On-Module. This is a flash module that's designed to replace a spinning-platter hard drive completely, geared towards small embedded systems and notebooks. I didn't really want or need a large HD replacement, so I found an 8Gb Transcend DOM unit that's absolutely tiny, and snaps directly into the IDE motherboard socket:
Again, it installed very easily under XP Pro, but wouldn't let me format it as NTFS, so it, too, stays as FAT32. I mounted it on the Secondary IDE channel, and since it's occupying the socket, I won't be able to add any more IDE devices to that chain without some sort of cable male/female Y-adapter of sorts. That's ok because my DVD burner is the slave device on the Primary IDE channel, no problem. The rest of the hard drives in this computer are SCSI, so we're safe. I noticed that the little Transcend DOM drive has a miniature selector switch for Master/Slave, so they know some folks will indeed use such an adapter cable. There were no instructions in the box with the drive, but I would imagine few would be needed.
This one performs the same way as the CF/IDE adapter combination, albeit with 8Gb of space vs. 4Gb. I've left just the pagefile on this drive, too, so that wear leveling has a chance. That, and I will bump up the system memory eventually, which will automatically expand the pagefile size.
So far, so good. About the only reason I tried this is because people recommended against it, but since SSD technology is finally established, I want to see just how long such an implementation would hold up in day-to-day operation. It may be robust enough to actually work, or it may crap out - we'll see.