The module works with session based saving. The more nodes to migrate, the slower it is. This is an exponential function. It makes migrating large sets of nodes a neverending job. We should think about workspace based saving/export/import/...

F.e. having about 10 thousands pages and focusing on "update node type" task, the migration is able to perform about 20 thousands of node types changes in first hour, 15 thousands in second hour, 10 thousands in third hour, ..., 3 thousands in 18th hour, etc. Such job would need about 3 days, which can't be acceptable. The linked migration.log is a good reference (covers less then a day), it was running on 8GB RAM and 2 CPUs.