Before a couple weeks ago, I'd agree that firewire 400 is absolutely better than USB 2.0. Then I started doing some actual testing and found that this isn't always true...
I own a pair of WD "My passport studio" portable drives. These are 500 GB portables with firewire 800, firewire 400 (via 800->400 cable), and USB 2.0. I've chosen not to install any WD drivers for these drives, and just use them "as-is."
My testing consists of creating a new time machine backup on the drive, and waiting until it's about 2/3rds of the way thru - then checking the throughput for writes via Activity Monitor. The machine isn't doing anything else while the backup is being performed. (The reason for waiting so long before watching on activity monitor is that Time Machine doesn't seem to kick into full speed writing until it's about 50-60% done..)
On an early 2008 blackbook (2.4 Ghz CPU and 7200 rpm internal HDD), with firewire 400, the writes max out at around 29MB/sec. With USB 2.0, I was maxing out at nearly the identical speed. (The same portable drive gives me over 50MB/sec when plugged into my MBP via firewire 800... so its not a drive limitation.) In other words, the drive was just as fast with USB as with firewire.
That being said, however, firewire is supposed to be able to maintain "full speed" data transfers in situations where USB devices start to lag behind - such as multiple attached devices or heavy CPU load on the hosting machine. I haven't actually tested this...
For an existing device that already has firewire - I'd use the firewire port as its supposed to be more consistent than USB. For raw speed, however, USB 2.0 can be just as fast as firewire 400 (and can actually burst to faster speeds - though very few controllers actually support that.)
If purchasing a NEW drive that would only be used for backing up, I'd probably choose a USB only drive, as they tend to be quite a bit less expensive than drives with both USB and firewire400. (For the MBP, I'll always pay extra for the firewire 800 speeds, though...)