Getting error message: "Captured packet too large" on many samples

Exactly the same issue for me and it finishes killed:

...
Captured packet too large: 24684 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 27580 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 18892 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 17444 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 17444 bytes, Max capture packet size: 16500 bytes
Killed

I run it on a smaller captured file (25M):

percona-playback --input-plugin=tcpdump --tcpdump-file=tcpdump_db01-2013-08-13_18-07.dump --tcpdump-mode=accurate --db-plugin=libmysqlclient --mysql-host=localhost --mysql-username=root --mysql-password=XXXX --mysql-schema=SSSSS
Database Plugin: libmysqlclient
 Running...
Captured packet too large: 17444 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 17444 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 18892 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 20340 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 21788 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 23236 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 24684 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 18892 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 29028 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 18892 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 21788 bytes, Max capture packet size: 16500 bytes
Captured packet too large: 20340 bytes, Max capture packet size: 16500 bytes
Error Report finished

Detailed Report
----------------
SELECTs : 0 queries (0 faster, 0 slower)
INSERTs : 0 queries (0 faster, 0 slower)
UPDATEs : 0 queries (0 faster, 0 slower)
DELETEs : 0 queries (0 faster, 0 slower)
REPLACEs : 0 queries (0 faster, 0 slower)
DROPs : 0 queries (0 faster, 0 slower)

Report
------
Executed 0 queries
Spent 00:00:00 executing queries versus an expected 00:00:00 time.
0 queries were quicker than expected, 0 were slower
A total of 0 queries had errors.
Expected 0 rows, got 0 (a difference of 0)
Number of queries where number of rows differed: 0.

Average of -nan queries per connection (0 connections).

Even if I force the capture size to 16500 bytes, it doesn't work:

# tcpdump -i any port 3306 -s 16500 -w tcpdump_$(hostname)-$(date +%Y-%m-%d_%H-%M).dump
tcpdump: listening on any, link-type LINUX_SLL (Linux cooked), capture size 16500 bytes
194229 packets captured
194256 packets received by filter
0 packets dropped by kernel

Seems like a pretty serious flaw. Has anyone found a workaround to this aside from not using tcpdump?