Build a PC for Linux development work, install two IDE drives and a SCSI controller with one drive and one CD/R. Choose reiserfs for your SCSI drive, to protect your work. You can significantly improve performance by using the entire, raw disk – partition table and all – as a solid reiserfs file system, so make the entire raw disk a solid file system. Bootsector, partition table, everything. Filesystem.
This works, because your Linux calls IDE drives “hdX” and scans them first, and SCSI drives “sdX” because it scans them second.
Now let your Linux update itself to a version that sees all drives as “sdX”. Notice this, and follow the wiki instructions to go on ahead and run grub-install before rebooting. Because, that’ll write a boot loader to the relevant boot drive. You booted of hda, hda becomes sda, so the boot drive should become sda. Read that particular bit of documentation again and be reassured that The Right Thing Will Happen based on the confidence expressed in the wording.
Where will grub write itself? To the bootsector. Where will the bootsector be? Oh, right about the same place your superblock is for the disk-wide filesystem that was previously on sda.
Postscript: Turns out that this bad idea was not so terrible; the reiserfs filesystem recovered, and it was part of a series of careful planning of my filesystems that allowed me to upgrade my box with a minimal disruption. Still. I think in future I’ll take the extra little performance loss and go with a partitioned disk: the performance hit isn’t really that big – an order of magnitude less than, say, running antivirus software.