I can't answer your question directly, but I have "heard" that CIFS is a problem. It works ok, when it is working, but when it hiccups, it is not easily corrected. When a CIFS connection is broken, it stays broken.
A vendor's engineer (Auspex) told me that he had found that an NFS client on NT actually provided better end results for all of the H-A storage systems that he had worked with. He suggested Samba or Hummingbird. NFS, he explained, keeps trying to reconnect until you fix the problem, and then NFS fixes itself.
NFS and CIFS have very different philosophies and design goals, and -- as a result -- they have different strengths and weaknesses.
The stateless NFS model is ideal in an environment where clients fail, servers fail, networks fail, and yet you still want to get some work done. To achieve this fault resiliency, the NFS designers were willing to make (minor) compromises in file system semantics.
The session based CIFS model is ideal when you want exactly the right file system semantics, especially in areas like file locking. The session based model also allows some cool performance optimizations (like opportunistic caching). Of course, the downside of a sessions is that when a session dies, you've got trouble.
So if you are trying to set up some kind of an fault resilient cluster, I can imagine that using NFS in an NT environment might actually make sense, assuming that NFS satisfies the application you are clustering, and assuming that you don't mind the administrative overhead of managing the PC-NFS software.
On the other hand, CIFS will probably do better if you've got lots of NT users running lots of NT applications. You'll get better file system semantics than NFS can offer, and hence less user confusion, and you won't have the management overhead of maintaining NFS on the NT clients.
Dave