I seem to remember reading in Atmel's literature that their parallel port programmer is dependent on software to generate the timing for the programming signals. To me that says that the programmer will operate differently on different PCs, plus, I think that relying on a PC to generate any form of short duration timing is a bad idea. I see three possible solutions.

1. Get a real programmer.
2. Build circuitry into your programmer to generate the timing. This could be done with a uC or logic. The programming software would probably have to be rewritten to wait for the programmer.
3. Get an adequately slow PC.

Cory