A middle ground does not appeal to me because simply knowing someone is present makes a huge difference and prompts a storm of buffs and search attempts from the would-be detectors.
What you're objecting to, however, is realistic behaviour.
Being able to detect signs that you're being stalked or followed without necessarily pinpointing exactly where the stalker is, and then going on guard, is much more realistic than the current set-up where you either can come within 20 of the stealther's score and are eventually guaranteed success; or your detection is useless. (I appreciate that Listen is more complicated, but nonetheless, the point stands that the range of values of conflict in detection vs stealth is effectively very small.)
Furthermore, the point directly leads to more mechanical freedom: if you increase the range at which detection might be useful even if not totally effective (say from 20 to 40), it also gives more freedom for devs to work with both for detection _and_ stealth. In other words: if detection can still be marginally useful if it comes within 40 of a stealther's score, then that means there's more freedom to push stealth scores up in general.
Frankly, the NWN stealth engine is bad. If we can find ideas to expand and improve it, let's do that, rather than stick with the status quo that swings on a knife's edge and thus encourages polarised and biased conversations.