you can also try "hciconfig hci0 reset" between stop and start but the "No rfcomm created" may be caused by something not directly related to bluetooth chip but the bluetooth software stack. Or maybe it is generic out of memory problem. Is there something interesting in kernel log (use 'dmesg') when bluetooth stops working or when you stop/start it?

Also maybe you somehow run out of rfcomm ports by allocating new ones without releasing old ones? Try rfcomm --help (bind, release command options)