I wan't to second this. I was using SoftwareSerial to talk to a serial LCD that is currently locked at 9600 baud (no documentation available). It seriously interfered with the stuff on Timer0 (millis() and delay()) and was causing my clock project to lose massive amounts of time due to lost Timer0 interrupts. I switched it to Paul's stuff here and it works perfectly now. There are no more lost interrupts and no more code running for more than 1mS with interrupts disabled (gag). His stuff works perfectly so far, but I"m not doing simultaneous send and receive. Since he uses interrupts and apparently the PWM feature for output, I expect it to have no trouble in that situation.