So a total database size of 3gb is fairly small.

With the affected rows to be roughly half of the database, that is about 1.5GB which is still pretty small.

Take into account a slower link to one of the replication partners and not wanting to overwhelm that pipe, I would batch process it from a server that is deemed to be the central/master server.

I would create a process that would delete 50k-100k records at a time. Put that in a sql agent job and then have it run once every 15minutes (in this case due to the slower wan link). This should be done relatively quickly, but the delay in between runs is just to give time to let the system process everything, clear buffers and breath.

You can tinker with the timings of the run since the 15min is just a starting point.