Sorry, I did not get the 2000 thing (hard to believe it's still out there so much 😉

Anyway, I'm not too familiar with DTS in terms of performance, even though technically I'm pretty sure the process would be fairly similar than SSIS. The idea I was having behind the RAW file export was to avoid a situation where you could lock both SQL Servers (if you use transactions for example). The RAW method allows you to do the import/export as two completely separate processes if you wanted. Some of the data from MS and other sites shows that RAW works really well, but I mean this is all dependent on your scenario/data too (http://consultingblogs.emc.com/jamiethomson/archive/2006/06/28/SSIS_3A00_-Comparing-performance-of-a-raw-file-against-a-recordset-destination.aspx).

As far replication is concerned, I'm definitely no expert in it, as we have tried only a couple of things with it where I work so I would not really be able to tell you. How would you deal with the deletion part in that scenario?

Maybe you could run a test load somewhere with replication and DTS and compare the two? Even throw in T-SQL as a third benchmark?