Here is what I found via one of the search links. It explains what CacheFS is
http://lwn.net/Articles/100330/
I see CacheFS and TSM working together on a temporary basis, on selected TSM Clients. But at the same time, there must be a common point of entry on a permanent server or a SAN LUN or volume group which permanently contains the data for the clients to gain access and utilize. This is where I would concentrate my TSM backups against.
Virtual mount points is a good option, but you are also going to have to consider or at least anticipate the amount of errors the activity log may generate if the include/exclude or cloptsets are scanning for the "mounted" filesystem and they cannot be found because they were not mounted or unmounted during a TSM session.
I see presched command here, looking for these mount points, continue if present, skip otherwise.
What is the purpose of this connection and the TSM backup/archive? If your scenario matches the description in the link, I would register the node on the TSM server as you normally would any client. Configure the dsm.sys to use virtual mount points and nodename options. Definitely use a pre and post sched command and insure logging is also configured so you have something to go by as a Tshooting aide when trouble strikes.
Since I am moving to a Solaris Shop environment next week, I will keep this in mind and get back with you once I am settled, that is if you have time.
Good luck, Hope this helps
Steven