Half-stroking sounds like an interesting idea, I'm not sure if it makes much of a difference if you keep your partitions (and the MFTs!) defragmented, though? I do have my 74gig raptor partitioned: 16gig for windows+apps, 4gig for source/docs, 50gig for things like games, "scratchpad", etc.
You are correct. As the drive gets fuller (more full?) this half-stroking business pays less and less of a dividend as the head moves farther afield. Fragmentation accelerates this effect.
And partitioning a big drive up like you have is the way to go (IMHO) but if more than one partition is accessed simultaneously any half-stroking benefits are lost.
In my case, I have a pair of 500MB 7200RPM drives RAIDed together (RAID 0) via a PCi controller made by SII. It's broken up into two partitions: one gigantic 900GB block for storing media & backups, and the other a 50GB block set up for temp files and miscellaneous transitory things. I have my Windows temp file located there, as well as a 2GB pagefile.
In this case, so long as I'm not spooling MP3s or a movie from the big partition, the read-head stays entirely in the small partition, travelling only a wee distance. Call it: 1/10th stroking.
This sounds wonderful in theory, but I'm not sure the experiment is working as intended. For instance, I usually *am* listening to music or watching a movie while I work, so the big partition gets accessed often.
Also, I have a huge-ass amount of RAM devoted to file caching (SuperCache II) so the software is mediating disk access and (I think) evening out the physical accesses. I mean, some temporary file operations generate NO disk activity, where I would expect to hear the drive going nuts.
I'm not sure the juice is worth the squeeze, though now that it's all set up I'm loathe to rip it all apart and redo it. It's definitely the fastest drive setup I've had so far, and love it -- but I think I could have chosen the partition sizes at random and gotten similar results.
Oh, and the punch line: the two 500GB 7200 RPM drives (RAID 0) replaced an aging pair of 35GB Raptors (RAID 0). I can detect no difference in performance or responsiveness, though HDTune seems to think the new array is slower.