I did some measurements on the NEC remote pulses
I used *20+16 for width
These are the results on average of 50-100 measurements:
Theoretical (usec): 560 1690 2250 4500
Measured (usec): 532 1674 2227 4549
Difference (usec): -28 -16 -23 +49
So *20+16 seems to work (Arduino 0021 and Duemilanove 168)
One additional comment: The reason I needed a bailout is because I am measuring HIGH pulses from the IR receiver. The normal off state of the receiver (no IR received) is HIGH, so if you get a spurious signal of one pulse the receiver out signal will go low, then high and stay there until the next event. In the meantime, pulseIn will wait for the end of the HIGH pulse forever...