Hi Deepak,
What Paul is saying is that you need to have the PST Migrator task on the same server where the user has their storage service and vault store, otherwise it will not know where to store the data.....so PST migrator is kind of tied in with the user's storage service.....but an archive task does not have the same restriction.
For example when customers have remote sites with exchange servers out there, they may choose to put an Enterprise Vault server in one of those sites and this server will run only the archive task....each night when the archiving schedule runs it will pass the data back to the central EV server that is running the storage service and indexing etc (the backend tasks). This is quite effective as the EV server at the remote site passes the data back via msmq which has about 50% less overhead than Mapi.
You could not put the PST migrator task on a server and then have it pass back the data to another server that is running the storage and indexing service....the pst migrator task needs to be on the same server as the storage service and the user's vault store.
What Paul has suggested with the 3 servers is definitely viable but only if you have these 3000 users split out over 3 different Exchange servers or more - that is because one EV server can archive from multiple Exchange servers but multiple EV servers cannot archive from one Exchange server. So you cannot have 2 archive tasks located on different EV servers targetting the same Exchange server, it would not work and it is not supported. But you can have the archive task run on a different server to the pst migration task as I mentioned previously.
So if all your 3000 users are on one Exchange server then I recommend that you get yourself a well specced EV server that does both the mailbox archiving and PST migration. No point splitting it out over 2 servers in my opinion because even if you set up 2 EV servers and 1 ran the backend services (so storage, indexing and the PST Migrator task) and the other EV server ran the archive task only, you would still end up with the backend server doing the bulk of the work (like the indexing and so on).
I hope this makes sense - if someone else wants to add any more details please feel free
cheers
andra