The fact that the Seed method ran only when the database changed was quite limiting for the database initializers that shipped in EF 4.1. It was limiting because sometimes you needed to update seed data without changing the database, but to make that happen you had to artificially make it seem like the database had changed.
With Migrations the use of Seed became a bit different because it could no longer be assumed that the database was starting empty–that’s kind of the point of Migrations after all. So a Seed method in Migrations has to assume that the database exists and may already have data in it, but that data might need to be updated to take into account the changes made to the database for the Migrations. Hence the use of AddOrUpdate.
So now we have a situation where Seed must be written to take account of existing data, which means that there is really no need to perpetuate the limitations of the EF 4.1 Seed method such that you would have to make it seem like the database had changed just in order to get Seed to run. Hence Seed now runs every time the context is used for the first time in the app domain. This shouldn’t change the way Seed is imlemented since it needs to handle the case where the data is already present anyway.
If it causes perf issues because you have a lot of Seed data then it is usually quiet easy to add checks into the Seed method that query the database to determine how much work needs to be done before doing it.