I think this might be the cause of a problem I'm noticing too.
I'm using an Huawei E160 3G USB modem on the RPi. PPP etc. works as I expect. But at random intervals, PPP says that the modem "hung up" (which it didn't, looking at the logs of the data) and cuts the connection.
Via minicom, AT commands never have a problem and the device works perfectly (in the same physical position) on a laptop with the same PPP / chatscript setup and NEVER hangs up. Literally the only thing different is the USB port that the other end of the cable plugs into (the modem + the cable it's connected to do not move one bit when swapping between the RPi and another Linux device). The signal quality at all times is top-notch and I've never experienced any problems with this modem before.
It seems to be more common the more I "stress" the USB connection, e.g. under GPRS mode, it hardly happens at all and I can ping google and download at a blazing speed (!) all day long, in 3G mode it connects fine every time, pings fine and consistently but if I try to do "traffic" (i.e. even a simple download of Google's homepage), it will usually succeed but almost always ends up triggering a disconnect during or soon after (presumably because the USB connection is losing packets as described).
I have so far eliminated account issues (it's a PAYG 3G stick and I'm paid up for today and it always works fine whenever I switch back to a normal laptop), PPP settings (identical config on both computers), IP issues (same IP given each time, with same DNS etc.), power issues, there are no intermediate hubs to cause a problem, there are no other processes running or trying to open the serial port (had gammu loaded once but uninstalled it while troubleshooting - and gammu had no problems sitting on the device all day issuing AT commands to check for new SMS, etc.), nothing else running, very idle machine (everything described done from the CLI, not the GUI), etc.
Power supply is fine (the one supplied by RS! But have also tried 2A adaptor that provides stable 5v by multimeter on TP1&2 and same problem). I took to using a small (6 inch) USB extension lead because I thought the problem might be related to the proximity of the USB ports (my keyboard port was leant on by the 3G modem when inserted next to each other), but having that extension has made no difference to either the RPi or the other laptop connection either, but has at least allowed me to "fix" the 3G modem and its connector so I can eliminate signal variability as much as possible.
I've set up hundreds of Linux PPP and modem connections in the past for everything from 9600 modem-to-modem, Hylafax servers with multiple lines, 56k modems for Internet sharing (from a custom single-floppy Linux), PPPoA/PPPoE, 3G modems, SMS daemons, etc. and I can't see a problem that I can narrow down to software or anything except the RPi USB connection at all.
Bloody shame because reliable SMS communication and the ability to report back over a 3G connection at will figure quite highly on the list of requirements for my first big RPi test project. Just trying to get on Google gives you (barely) the front page before it just dies and PPP kills the connection and tries to redial.