There are some very solid, and pretty intuitive, ways to speed up Merge.

For my cross-server ETL, I set up a bulk-copy database that just copies the source tables from the other server. Very fast because it's minimally logged and doesn't incure DTC locks. Then I Merge from that. But that kind of "stage first, load second" isn't specific to Merge. I'd have done that no matter what my final choice was for loading the target tables, since it would apply equally to kill'n'fill, or update/insert (2-step), or anything else.

Another way to really speed up Merge is to make sure the On statement from source to target, is properly indexed, and (preferably) has index-enforced uniqueness for the source. For complex source queries, it can be faster to load a temp table, index that, then use Merge with that as the source.