Great article Adam.
Regarding cross server merging. Isn't it best to use a ETL tool for ETL tasks, making use of in-memory caching, instead of a linked server?
Regarding checksum comparisons how do you handle collisions. I.E. the "Birthday Paradox"? I've used 'md5' or 'sha*' in the past but never checksum because of the increased collision probability. Do you have code that resolves collisions?