Wow thanks for all the responses.
MikeMl- re airspeed pitot tube; probably not usable because of the massive air pressure pulses from the unsilenced exhaust. With 1000hp and nitro methanol etc just standing in the vicinity of the engine tugs your shirt around all over the place! Then there would be windage from the blower belt and spinning back tyre. Also the ground wind on the track will be gusty due to the concrete barriers etc and hard to compensate out.
Pommie- re accelerometer, the datalogger has one in that does an ok job, barely, the trouble is the intial forward launch is usually accompanied by bucking twsiting etc and intermittant traction of the tyre it's not much use for the resolution needed (ie 100 samples per second preferred) as short term data is so extreme and noisy.
Jbeng- A radar gun is an interesting idea. It would need to be able to get a doppler reading from the ground itself, at high sample rate. I'm not sure how practical it would be or how reliable in such a high vibration environment.
Crutschow- re the mouse sensor it still seems the better option. The problem looks to be focus, there's no part of the bike at a stable distance from the ground at the launch. And any type of sprung lever to keep the sensor at the right ground height may as well just read the sprung wheel rpm.
Rbecket- Very interesting idea re the contact audio sensor. I think it would work ok with processing as the noise generated would be proportional to surface speed. But practically it's a bust as the track has different surfaces including a solid layer of melted rubber and again you would need to keep it in contact with the track... The track itself is reasonably well setup with laser sensors and automatic printouts of the time and some launch times etc. It's just for critical tuning of the clutch and tyre setup etc.
At this point I think I will tell him to make a sprung lever with a wheel on it, and we can use a hall sensor against a fine cog rotor on the wheel to get fast feedback. It's clumsy and large but he can unbolt it easy enough when not doing tuning runs.
I'll also look into Pommies idea, maybe a custom accelerometer can be made with hardware or math damping to do a pretty good job. The instantaneous ground speed is probably less important than the actual forward thrust anyway for the data he needs to set up the bike.
What a shame there's not an optical sensor that you can just point at the ground from 6" to 18" distance and get a good speed reading.