SPI writes are a great improvement over bit-banging Arduino ports. There are a couple of ways of accelerating SPI writes with the Arduino.

• Most example software uses digitalWrite() to set the chip select low and high.
• digitalWrite() has a lot of overhead and turns out to be pretty slow
• Directly accessing the Chip Select bit can speed up writes dramatically.
• Direct access requires you know the register and bit that the port is connected to.
• For instance, for an ATMega32U4 based design the SS line is D17 on the Arduino or PB0 on the Atmega32U4.
• To directly set the PB0 bit:
  

• PORTB &= 0xFE;  // Directly set the Chip Select line low

• PORTB |= 0x01;   // Directly set the Chip Select line high

• The SPI clock register (SPR1 and SPR0) and the SPI2X bits can be set to faster speeds.

• There is a warning in the datasheet that the SPI needs to be run at fosc/4 when used in slave mode (N/A in this case).
• The fastest rate that the ATMega32U4 allows is half the processor clock.
• For a 16 Mhz processor, that's 8 Mhz which is a 125 nS clock rate.
• In comparison, an AD9833 DDS chip can run with a 40 nS clock min (25 MHz max or 3X the speed of the Arduino SPI).
  
• To set the clock to the higher rate:
  

• SPI.setClockDivider(SPI_CLOCK_DIV2); // Set the SPI clock to 8 MHz rate

Measurements - 16-bit Transfers

Resulting speed for a 16-bit transfer is 3.26 uS from chip select low to chip select high.

Top Trace = SCK     Middle Trace = MOSI (Data)           Bottom Trace = SS (chip select)

Cycle Time (chip select low to chip select low) is 3.64 uS.

The Front edge (chip select falling) timing is:

The back edge timing is:

Top Trace = SCK     Middle Trace = MOSI (Data)           Bottom Trace = SS (chip select)

Links

• Timing the shiftOut() function