Hour long delay between each large DLE during dump

I have been using Amanda to backup 2-3 TB of disk to a Quantum tape
library SDLT600.
The data does get backed up, but, after each DLE, there seems to be a
very long pause (hour or more) before the next DLE starts backing up.
 In addition, the load on the system is above 4.  From top -ci:

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
 5739 root      17   0     0    0    0 R 99.9  0.0  13074:45 [jfsCommit]
21176 amanda    18   0  9448 7976 7260 R  3.7  0.4 467:43.62 taper conf
17646 amanda    16   0  2056 1020  752 R  0.3  0.0   0:00.53 top -ci
    4 root      RT   0     0    0    0 R  0.0  0.0   0:04.50 [migration/1]
    7 root      10  -5     0    0    0 R  0.0  0.0   0:29.73 [events/1]


The system is quite responsive, but I cannot determine why the pauses
occur.  The log file shows that the pause occurs after the last CHUNK
is processed for a DLE, and the CHUNKSUCCESS is completed.
For example:

>>>Amanda log file>>>

CHUNK taper host entry...
CHUNK taper host entry...
CHUNK taper host entry...
CHUNK taper host entry...
SUCCESS dumper host entry
CHUNK taper ost entry
CHUNKSUCCESS taper host entry

<<<HUGE Delay here for hour or so, with corresponding high load>>>

INFO taper mmap failed (Cannot allocate memory): using fallback split
size of 512000kb to buffer host:enrty in-memory
CHUNK taper host entry ...

<<<End Log file<<<


This doesn't seem to occur for small DLE's, like /etc, but does for
the larger 100+GB DLE's.

The system is under very light load during the backup, and the backups
are usually going directly to tape, since I don't have a holding disk
large enough (at least one that is smaller than 4TB, which is a limit
that I ran into for holding disks).
The tape speed is sufficient for me.  12-18 MB/sec
The backup speed and delay is problematic, since it pushes my weekend
backup into Monday morning.

Any help would be appreciated.