There is indeed a delay. What's happening is this:
1. Press a key, the key turns on
2. Wait 100ms
3. Sample the absolute value of difference between the two key sensors (one on top and one on bottom of each key) — this will be the intial "comparison point".
4. Wait 250ms
5. Sample the value again and compare to the initial comparison point
6. If tilt threshold is crossed, engage tilt — else, continue sampling and comparing.
So you are correct. At the absolute minimum there is 350ms of delay (0.35 seconds) between the key turning on and tilt engaging.
I doubt this will be changed though. We went back and forth many, many times trying to settle on a tilt behavior. When it was really quick and easy to engage tilt people would say it was too difficult to play chords in tune (when pitch bend is mapped to tilt). When it was more difficult to engage tilt people would say it was too hard to use. Eventually we settled on the current method.
As cool as it would be for a person to master playing a QuNexus with insanely sensitive tilt mapped to pitch bend (like some weird violin with a keyboard layout), making tilt super easy to engage will disappoint more people that it would help.
I can put in a feature request for a user-settable delay time though. But honestly I think our editors could use some paring down of options, rather than adding more.
2, 3, 5, 13, 89, 233, 1597, 28657, 514229, 433494437, 2971215073