2 things that will effect the performance here...
1) millions of files always perform worse than anything else, it's just simple physics with disks.
2) Running out of inodes shouldn't cause any performance impact. You may find that as you run out of space you get some sort of performance degradation (as the system needs to seek for free blocks longer). I would expect that the performance impact would be more likely because there are millions of files, and clients may have to enumerate these as they are browsing.
Unless you really need it, turn on "no_atime_update" on the volume (vol options volname no_atime_update on). This stops the filer modifying the access times of files, and can be quite a big performance boost with larger file counts. To be honest, I often do this with a lot of CIFS / NFS volumes anyway as I have yet to find a really good use for access times.
It also depends how busy your filer is now. If your filer is comfortably ticking away, you may not notice any impact. If your filer is already busy, you may notice it. But I would expect this to be more from having millions of files, and less because of a lack of inodes. The system wouldn't need to seek for free inodes as it would do for free blocks on a disk.