asterisk users - Jul 2005 - Dropped calls if transferred across servers into MeetMe with mobile source

I have an application where calls come into an *box from a DID  
provider, and may be transferred to a meetme conference on another  
*box (the call is released by the first *box after transfer).

These are ulaw IAX channel calls, and if the source is from a Verizon  
or Nextel mobile phone to the DID (other carriers not tested), the  
call drops about 2-3 minutes after it joined the meetme conference.   
POTS originated calls do fine - they do not drop.  I've reproduced  
this consistently, and across two different DID termination providers  
and several different mobile phones.

I'm seeing the behavior on 1.0.7 and 1.0.9.  Calls don't fully drop.   
Meetme shows a reduction in the participant count, and the conference  
exit tone plays, but the mobile phone thinks it is still connected...  
AND other call participants can still hear the 'dropped' person, but  
that person can't hear anything.

Also, if I change the first *box iax.conf to notransfer=yes, all  
calls are reliable (but, of course, I'm tying up resources...not a  
good long term solution).

Console output is as follows for problem calls:
Jul  5 15:18:44 WARNING[9256]: chan_iax2.c:1477 attempt_transmit: Max  
retries exceeded to host xx.xx.xx.xx on IAX2/yyy@ xx.xx.xx.xx:4569/3  
(type = 2, subclass = 4, ts=65540, seqno=1)
Jul  5 15:18:44 WARNING[9256]: app_meetme.c:962 conf_run: Unable to  
write frame to channel: No child processes
   == Spawn extension (toll, 1001074, 5) exited non-zero on 'IAX2/ 
yyy@ xx.xx.xx.xx:4569/3'
     -- Hungup 'IAX2/yyy@ xx.xx.xx.xx:4569/3'

I've experimented with jitterbuffer on and off, different qos  
settings (including high reliability), and different meetme options.   
I haven't been able to impact this behavior.  There is an agi that  
executes when the call arrives at the meetme *box (before meetme is  
joined). It just hits a db, sets some variable values, and exits  
cleanly -  and again, it's not until 2-3 minutes later that I see the  
problem, and I don't have any problem with POTs sourced calls.

The big variable seems to be whether the call originated from a cell  
phone or not, and that it was transferred to a second server.

This is really strange, and I've even pulled in someone else that  
does Asterisk work just to do a sanity check and make sure I wasn't  
missing something obvious... no such luck.

Any thoughts or insights?