The early Raspberry PI (RPI) boards had polyfuses on the USB ports. They are electronic resettable 140 mA fuses. The problem is that they do not have zero ohm resistance. This means that by increasing the power drawn from the USB port the voltage drop across the polyfuse increases. By drawing approx. 100 mA and more from the USB port the voltage drops bellow the value allowed by the USB specification and the connected USB devices may behave incorrectly. I encountered this when I was backporting the R820T tuner driver to the kernel 3.9 (needed for some RTL2832 based DVB-T dongles). Some of the DVB-T cards I had worked OK, some exhibited random failures during I2C writes to the R820T registers which resulted in occasional kernel panics.
You can workaround this by using a USB powered hub or you can fix the RPI PCB. The fix is simple - just bypass the F1 and F2 polyfuses. This fix is already implemented on the later rev. 1 and rev. 2 boards, so the later boards don't exhibit this problem. You can remove the polyfuses and short the pins, but an easier approach is to leave the polyfuses on the PCB and just short them out by wire or just by tin. You will still have the protection from the main F3 polyfuse (0.7 A). The other advantage of this mod is that it will allow you to power the RPI from the USB hubs / devices that provide power upstream.
The F1 and F2 polyfuses are green (at least on my RPI boards) and are located near the status LEDs. This modification is applicable only to the early rev. 1 boards. They probably have the HW revision code 2 or less. You can get the HW revision code by running:cat /proc/cpuinfoIf your HW revision code is 3 or above this problem is already fixed on your board.