I have had problems running coda over Mobile IP and IPsec on wireless
lans.  I believe my problems were due to fragmentation; the rpc2 code
will piggyback some amount of side-effect data onto rpc packets (2900
bytes?).  This is 2 Ethernet packets when fragmented.  However, if
each of these then gets put in ESP and needs additional fragmentation
due to being big by the size of the ESP header, you might have 4
back-to-back packets needing to all arrive to get reassembled.

I'd recommend turning the MTU down on interfaces heading towards an
IPSEC tunnel to 1450 or so from 1500.
I would also recommend watching all the relevant interfaces with
tcpdump, paying close attention to fragmentation.

Also, try the following patch to rpc2, which reduces fragmentation.
(I haven't checked it for ESP plus mobile IP, but I think the 1400 is
the max body bytes of the header plus data, so adding
 2 * IP  + 1 * ESP + 1 * UDP
should be ok, since that's around 60, which is < 100.

Index: rpc2-src/sftp.h
===================================================================
RCS file: /coda-src/rpc2/rpc2-src/sftp.h,v
retrieving revision 4.22
diff -u -r4.22 sftp.h
--- sftp.h	2000/01/31 22:23:18	4.22
+++ sftp.h	2000/05/01 11:48:45
@@ -55,7 +55,7 @@
                                 /* Changed from 2 on 27 Feb 1997 by bnoble */
 
 /* (header+body) of largest sftp packet (2 IP fragments on Ether) */
-#define SFTP_MAXPACKETSIZE	2900
+#define SFTP_MAXPACKETSIZE	1400 /* 2900 */
 #define SFTP_MAXBODYSIZE	(SFTP_MAXPACKETSIZE - \
 				 sizeof(struct RPC2_PacketHeader))
 


        Greg Troxel <gdt_at_ir.bbn.com>