Don't foget that the speed of sound changes as your altitude goes higher.
Anyone here from Denver?

Also, if we alllow a minimum of 1ms 'delay' and increment in 1ms jumps then that allows 300mm or about 1 foot of delay per increment. You could halve that by allowing .5 of a ms increments - that would give a six inch granularity at sea level anyway - you would need 3 digits on the display - 1 for the units and one for the dot and one for the '5' or '0' of the increment. If you allowed for 20 ms delay then you would need 2 digits for the units rather than 1.

Even with .5 ms increments it would only require 40 'clicks' of the knob to go from full right to full left delay assuming a 10 ms 'in car' delay range and one click of the knob per increment - I am not sure how many clicks the knob does per turn. probably 16 or 32.

Don't forget though that as I pointed out earlier, some of use may use this inside our homes, and its not out of the question and that more than 10ms worth of delay may be required.

I think 20ms delay on the left or right channel would be heaps for any situation I can imagine.

The big question is, how much delay can we get without using up too much precious kernel memory - I assume you will have to store the delayed channel samples, there would be 88.2 bytes per ms of delay (for a single channel), so 20 ms would be 1700 bytes of buffer required.
This buffer may already be available in the Voladj hack Richard did - the PCM data stream certainly is. How far ahead does Richard's hack look for the Voladj stuff?