In this post on Microchip forum, xiaofan shows libusb code for HID which works successfully.
http://www.microchip.com/forums/tm.aspx?m=340898

This problem doesn't reveal on the most of drivers on Windows and MacOS X.
- These drivers put Set_Configuration just once after bus reset.
- These drivers don't put Set_Interface, unless target interface have any alternate.

But libusb code (both on Linux and Win32) may put Set_Configuration twice or more, without preceding bus reset. Also, libusb code may put Set_Interface at any time.

Here is a fix of Set_Configuration handler.

C:\SiLabs\MCU\Examples\C8051F34x\USB_HID\HIDtoUARTExample\F3xx_USB0_Standard_Requests.c

void Set_Configuration (void)          // This routine allows host to change
{                                      // current device configuration value
   // Device must be addressed before configured
   if ((USB0_STATE == DEV_DEFAULT) ||
   // and request recipient must be the device
   (SETUP.bmRequestType != IN_DEVICE) ||
   // the index and length words must be zero
   SETUP.wIndex.c[MSB]  || SETUP.wIndex.c[LSB]||
   SETUP.wLength.c[MSB] || SETUP.wLength.c[LSB] ||
   SETUP.wValue.c[MSB]  || (SETUP.wValue.c[LSB] > 1))
   // This software only supports config = 0,1
   {
      Force_Stall ();                  // Send stall if SETUP data is invalid
   }
   else
   {
      if (SETUP.wValue.c[LSB] > 0)     // Any positive configuration request
      {                                // results in configuration being set
                                       // to 1
         USB0_STATE = DEV_CONFIGURED;
         EP_STATUS[1] = EP_IDLE;       // Set endpoint status to idle (enabled)
         POLL_WRITE_BYTE (INDEX, 2);   // Change index to endpoint 1
         // Set DIRSEL to indicate endpoint 1 is IN/OUT
         POLL_WRITE_BYTE (EINCSR2, rbInSPLIT);
         POLL_WRITE_BYTE (EINCSR1, rbInCLRDT | rbInFLUSH);   // clear data toggle and flush FIFO<------------ add this line
         // Enable double buffering on out endpoint
         POLL_WRITE_BYTE (EOUTCSR2, rbOutDBOEN);

         POLL_WRITE_BYTE (EOUTCSR1, rbOutCLRDT | rbOutFLUSH);   // clear data toggle and flush FIFO    <------------ add these lines
         {                                           // clear double buffer
            BYTE eoutcsrl;
            do {                                     // wait until FLUSH complete
                POLL_READ_BYTE( EOUTCSR1, eoutcsrl );
            } while ( eoutcsrl & rbOutFLUSH );
            POLL_WRITE_BYTE( EOUTCSR1, rbOutFLUSH ); // clear double buffer
         }
         POLL_WRITE_BYTE (INDEX, 0);   // Set index back to endpoint 0
         Handle_In2();
      }

As of Set_Interface,
HID interface doesn't have any alternate interface. Then, the firmware can reject this request at all.

C:\SiLabs\MCU\Examples\C8051F34x\USB_HID\HIDtoUARTExample\F3xx_USB0_Standard_Requests.c

void Set_Interface (void)
{
   Force_Stall ();
/*
   // Make sure request is directed at interface
   if ((SETUP.bmRequestType != IN_INTERFACE)  ||
   // and all other packet values are set to zero
   SETUP.wLength.c[MSB] ||SETUP.wLength.c[LSB]||
   SETUP.wValue.c[MSB]  ||SETUP.wValue.c[LSB] ||
   SETUP.wIndex.c[MSB]  ||SETUP.wIndex.c[LSB])
   {
      Force_Stall ();                  // Othewise send a stall to host
   }
   if (EP_STATUS[0] != EP_STALL)
   {
      POLL_WRITE_BYTE (E0CSR, (rbSOPRDY | rbDATAEND));
                                       // Indicate SETUP packet has been
                                       // serviced
   }
*/
}

Tsuneo

[Revised]
July 30, 2010
Added data toggle clear for OUT endpoint, too, on Set_Configuration handler
Added FIFO flush to IN and OUT endpoints on Set_Configuration handler

I added these changes to the Set Configuration and Set Interface routines but they didn't fix the problem I am seeing.