You can, besides writing grldr.mbr to the MBR or making a VBR invoking grldr (which is - I presume - what you mean by "installing") load it from BOTH NTLDR (which won't be used in a PE 2/3/4/5) and BOOTMGR (which is needed for a PE 2/3/4/5).
The "installation" (which is NOT at all an installation) consists in two steps:
- copy to the root of the device a grldr (and it's menu.lst)
- create in the root of the device a file named BOOT.INI with these contents:
[boot loader] Timeout=30 default=C:\grldr [operating systems] C:\grldr="grub4dos"Of course you are perfectly free to call the above procedure "installation" and to consider it "difficult" or "advanced", or "another level of complexity".
Same goes for "differential testing", no matter what you believe (and how much right you are in your beliefs) the differences between running a tool in a "full" OS vs. running it in a PE (as long as it actually works in the more limited PE environment) won't turn a turtle by magic in to a hare.
For the early classification between "slowish" and "fastish" we could use even VM's, like Qemu (to be able to test a "pure" IDE interface), or another one (to be able to test a simulated SCSI in VMWare or Virtualbox for a SATA one) with a "full" OS installed in it, since we are going to measure "relative" differences in speed, it will be as effective as that.
BTW, what I am telling you is that - at least in this preliminary phase - it makes NO sense whatsoever to attempt imaging anything bigger that (say) 2 times the size of the cache of the hard disk, and definitely attempting to image a 1.5 Tb disk would be foolish and unneeded.