This is a bit of an onerous process, but given the amount of problems we are finding I think its necessary.

Go through the all the code in siteregistration.php - at any point where the process can fail, insert debugging() calls with relevant information about the site which is attempting it and what is causing it. The goal here is to apply this to the production site and have that debug info go to the error log so we can analyse what the heck is going wrong when we get reports.