The only length whip that "works" without any matching devices is a quarter-wave, which is around 50 Ohms or so all by itself (with a good, and solidly connected ground plane). But more gain can be achieved by going to larger aperture antennas, which unfortunately aren't 50 Ohms impedance. 5/8-wavelength was settled on many years ago as probably the perfect compromise between performance and achieving a reasonable match using low-loss components, and has become the de facto standard for VHF whip antennas over the years for good reason. But you cannot directly feed a 5/8-wave whip, as it's not 50 Ohms, and has capacitive reactance. A series matching coil adding equivalent inductive reactance is all that's needed to make the whip match 50 Ohm coax very well, and if that coil is large in diameter and made of very low-loss material, it will introduce no measurable loss at all.
This is what Larsen achieves with the NMO-150 coil, as opposed to many others on the market. The coil is silver-plated copper with very high Q and excellent L:D (length-to-diameter) which helps provide the high Q. It doesn't really introduce any loss at all.
The reason a through-hole mount works better is simply reduced ground losses: The lower the ground resistance, the more power is radiated by the antenna and not dissipated in the signal return path (ground). It's really as simple as that. No mag mount can achieve this, although very good ones can come pretty close. Ironically, about the best mag-mount base I've ever tested is also from Larsen (it's rectangular and ugly, but has nearly optimum coupling capacitance through its foil base and the car's paint, and works very well on 2m).
The reason the middle of the car roof works better than a trunk lip is because the center of the roof is the highest location on the vehicle, has increased LOS and thus increased OTH range via tropospheric scatter, and isn't blocked by other sheet metal parts of the vehicle -- or other vehicles that surround you. Unless you're driving a 12' tall car, you'll still be occasionally blocked by other, taller vehicles (mostly trucks, buses, etc), but at least with a whip in the center of the roof, you're not continually blocking your own signal. And of course the center of the roof provides the most optimum ground plane for the antenna; a trunk lip provides a pretty good ground plane in one direction, but almost no ground plane in the other.
Actual empirical measurements indeed indicate an average 6dB difference in radiated field strength as you drive a car in a circle and measure the field at a point a 100 or so wavelengths away at similar elevation; this has been performed an awful lot of times by many experimenters...probably most notably Motorola, who did repeated studies like this back in the early 1960's and published all their results.
WB2WIK/6