I'm mostly concerned with the serial port, SPI port, I2C and PWM. Are they hardware, or bit-banged?
There are hardware and software implementations of serial communication and PWM, at least. SPI and I2C may be bit-bangable, but they are supported in hardware, for specific pins.
How much overhead does Sketch add?
Depends on the sketch, the OS (which typically defines the compiler in use), and the compiler optimizations selected (hard-coded, that is).
What is the instruction cycle time of the Atmel chips?
Depends on the clock speed, which is typically 16MHz, so each instruction takes 62.5 nano-seconds.
Or - if someone could point me to a (hopefully single) source for the above information, I would be most appreciative.
Well, that would be the ATMEL data sheets that you don't want to read.