I made programmer with optoisolation in connection USB and target controller on base of AVR-Doper. For connection parts used 4m #24AWG cable (LAN cable). Circuits and open code into archive, what go with letter. Work under OS XP: time of programming ATmega8 at frequency 480 kHz, code size 6.7 Kbit is 5.5 sec, where 3.3 sec time of programming and 2.18 sec time of verify. Work under OS Vista, notebook Toshiba L40: time of programming ATmega8 at frequency 480 kHz, code size 6.7 Kbit is 22 sec, where 13 sec time of programming and 9 sec time of verify. I think, what I will stay Linux on this notebook. I don’t widely test this design, because I have not possibility and time for it. I can say what ATmega8 and ATtiny2313 it programming at interface frequency 115 kHz and ATmega8 with clock 12MHz and interface frequency 480 kHz, but that is all. Also I tried down voltage to 3V with 0.1V step, threshold of work is 3.5V, may be it can work at 3.3V if down clock USART as minimum twice. 

	List of program changes, what I made:
1. Cut firmware AVR-Doper on two part at moment transfer TX and RX buffers and connection them over functions used USART interface;
2.  Adapted program of outboard part to work on events USART interface; 
3. Adapted connection direct to target controller, without driver;

Also I changed count timer in outboard part. Outboard have not interrupt for USB polling, because so kind count ticks is more robustly from overflow TCNT0, than calculate difference between new and old value TCNT0. And compiler isn’t always adequate work with mix of signed and unsigned values.
In file Vreg I wrote PI regulator, it can work more stable, than simple P regulator. I don’t test your circuit with this regulator, but in other applications this code snippet work good. There is only may be need tune coefficients.  

Sergey Merkulov.