1)  Partition for speed

Aim for 4k clusters, which FAT32 will give you up to 8G.  So I get the
biggest HD that isn't silly-priced, set up a 7.99G FAT32 C:, and keep
only Windows, core code, swap and temp files there.  The rest of the
HD I set as an extended partition with thrre logical volumes on it; a
2G FAT16 D: for small data files, a huuuge FAT32 for "everything else"
(games, suspect files, music, pictures, movies etc.) and 2G FAT16 F:
for "cold storage" (pre-installs, Win9x .cab set, backups).

If you have two physical HDs, and the second is about as fast as the
first, then (and only then) does it make sense to locate the swap (and
perhaps temp) files off C:, on the second drive.

2)  Kill off underfootware

http://users.iafrica.com/c/cq/cquirke/startup.htm refers.  In
particular, I kill off: Fast Find and other useless indexers, Outlook
journalling, System Restore (in WinME), commercial malware in general,
and any SysTray chorbs that accompany apps that I use once in a while
(QuickTime, Real gunk, WinAmp, etc. etc.)

Use System Monitor to see how much swap file is in use - that's the
only metric I find useful for assessing when adding more RAM will pay
off (as it will, if a fresh startup already uses swap).

3)  Shrink IE's dumb-a.s fat cache

By dufault, IE hogs something crazy like 256M+ of HD space for
yesterdays (yesteryears?) web pages.  Stomp that down to 20M or so.
If your internet connection is fast enough to pull down 100M of gunk
within a week, you don't need caching to speed it up.

4)  Keep large cold lumps out of C:

An extension of (1).  A lean C: means shorter head travel and faster
disk maintenance (i.e. Scandisk and Defrag).  The idea is to keep 90%
of HD activity (C:) or 95% (C: and D:) within the first 10% of the HD
(or better, if HD bigger) no matter how large your movie collection
etc. gets.  This is the single best way to maintain performance.

5)  When to Defrag

Defrag only when you know the HD is physically OK and the system is
stable - never defrag to attempt to fix a flaky system!  I defrag:
 - after I delete or uninstall a lot of stuff
 - before I install stuff
 - after a volume has filled up and recovered

The last needs explanation.  Usually, new files are created in the
large contiguous free space at the end of the HD, starting from the
beginning of this free space.  But if the volume has filled up, this
forces new files to be created in the "holes" between files, where old
files have been deleted - causing fragmentation.  If you then clear
off material to free up space, the new space created is also in patchy
holes.  A defrag is needed to bring existing files closer to the
"front" of the volume, and to consolidate free space after that.

These are general strategies with speed in mind, but usually I'm
thinking as much about reliability, data safety and "safe hex".  The
C:, D:, E: and F: strategy serves those needs as well.  

Consider: A PC that's unusable for days required to "just re-install
Windows" after some disaster or other is *really* slow for those days

Error Messages Are Your Friends