There is a limit of 1000 partitions per partition scheme and you can only partition on a single field, so if you intend to multi-tenant beyond 1000 instances you are going to have to jump through a lot more hoops. You can extend the limit by using a partitioned view on top of multiple partitioned tables, but this increases the management overhead. You can use the DMVs and create your own automated system that generates new partitions per client / tenant and manages the problem but it will be specific to your application and not generic.
At present there is no automatic dynamic partitioning in SQL Server, it was mentioned at the PDC09 in relation to the SQL Azure future roadmap, but I did not hear of it for SQL Server.
Your alternative choices are a database or SQL Instance per client, there are benefits to this approach in that you give yourself far more opportunity to scale out if the needed arises, and if you start looking at a larger data centre, you can start balancing the SQL Instances across a farm of servers etc. If you automatically have all the data in a single database.
Other things to take into consideration:
Security: Whilst you have the data in a single database with partitioning, you have no data protection. You risk exposing one client s data to another very trivially with any single bug in the code.
Upgrading: If all the clients access the same database, then the upgrades will be an all or nothing approach - you will not be able to easily migrate some users to a new version whilst leaving the others as they were.
Backups: You can make each partition occupy a separate file group and try manage the situation. Out of the box so to speak, every client s backups are mingled together. If a single client asks for a rollback to a given data you have to plan carefully in advance how that could be executed without affecting the other users of the system.