i make an md5 of the file on the recorder, and then use that md5 sig to test the one that is copied to the harddrive...
If you are using a firewire connection to access the recorder harddrive as a mounted block device, the data transfer should be through the asynchronous channel and error correction handled by the physical layer protocol should ensure that the upper layers see no bit errors. When I saw MD5 issues with the 722 file transfers way back when, I blamed the 722's buffer management and not the firewire link layer. But who really knows?
I assume that you are doing your MD5 test with the recorder mounted as a drive on your PC. If so, then the file data is being transfered once to your workstation for the initial digest calculation and then again when you do the transfer to the PC HD. You are using the same link twice. If your digests don't match, you can't be sure which is the correct one or even if either is correct. The only sure solution is to have a recorder that can internally calculate a digest and report that for comparison against the digest calculated on the file after it has been transferred.
It is a nice sanity test because a random error won't appear the same way twice but it does not ensure that the firmware isn't sending erroneous data in a repeatable way. If the device has a fault in the firewire driver, an internal digest calculation would reveal a problem.