Finally I have had the time to run vmstat during cache purge.
I have attached one file with vmstat data from our nfs server and one with data from the application node.
At this time the cache-directory was around 1.5G in size. The run timed out and it seems like the cache directory didn't really get purged at all this time.
As a whole our moodledata is 62G
59G -> moodledata/filedir
1.5G -> moodledata/cache
Memcached is probably the way to go... but I still wonder if our NFS-server is the problem in this case.