I have an issue on database (2 node RAC system) on AIX.  Our system is already around 25T and our full backup size almost 3,7Tb with using compress option.

Our number of datafiles are almost 1054.  We are taking our backup to disk using 16 channel. Total full backup duration is almost 4,5 hours.

Our backup duration became almost 7,8 hours last 3 days. I investigated backup log file and I noticed backup start 16 channels and that 16 channel backupsets job done almost 4:30 hours. After 16 channels backupsets  end,  one new channel opening and it keep taking backupsets by using 1 channel.

We investigate an issue with Oracle support. Here is the explanation of this  behavior

RMAN will try to balance the work load such that each channel processes the same amount of data.
17 backupsets are produced, each channel is processing 61-62 datafiles so you have ~ 1054 datafiles.
RMAN will initially take the no# files and divide by no# allocated channels – this gives 65.8.
The upper limit for filesperset is 64 so :1054/64 = 16.46 so rman will create 17 backupsets.
As you have 16 channels allocated the first 16 chs will process concurrently and the 17th backupset will inevitably run in isolation.

If you allocate 18 chs you will get filesperset of 59 and all 18 chs should process concurrently.

So We increased our number of allocate channels for backup and Issue has been resolved.