Via Inel, there's an interesting article about the amount of carbon released by fire relative to other sources in the US - it can be very substantial, about 4-6% of anthropogenic emissions.
The article acknowledges the conceptual problem that in the long-term, natural fire cycles are carbon neutral, with a constant cycle of burning and regrowth. I suppose the same is true about any natural carbon process, but the time scale for forests is just the right, multi-decadal length to complicate our plans to address greenhouse gases. Also, we don't have natural fire cycles any more, and the change can alter the amount of time carbon spends in the air instead of in a plant or in the ground.
Another wrinkle is with forestry offsets. You can't just measure the difference in carbon storage between a forest logged on a fast rotation cycle and a forest allowed to mature to old-growth stage, and say that's the offset amount. You can lose that offset in the next fire and have to wait for it to come back. It's not an insolvable issue, though. You just discount ("offset") the amount of carbon you calculate to have stored by the percentage of time you expect it will spend in the air over the long term due to fires. Here in California, with our quick-growing, 500-years-between-fires redwood forests, that discount won't have to be all that big.
I will concede this is another complication to using forestry offsets, despite my previous defense of the concept. Too bad, though - it's not like climate models are simple and error-free, and being complicated doesn't equal being useless. I say, "Viva conservatively-calculated forestry offsets!" Put that on a bumper sticker next to a picture of that idiot Che Guevara.