I've been hoping that ADTech would weigh-in on this topic... I don't have a DB8E in hand at this point, just a DB4E and 91XG. So, combining my first hand knowledge of those designs with the published photos of the DB8E, here goes...
The 91XG and DB4E both employ electrically balanced 'bow-tie' elements. To be a bit more technically accurate, the 'bow-tie' element is a form of a folded dipole
http://en.wikipedia.org/wiki/Folded_...#Folded_dipole, which inherently has a terminal impedance of roughly 300Ω at it's resonant frequency. The 91XG uses one directly connected and the the DB4E has four which are interconnected via balanced lines (the parallel heavy gauge wires running vertically on either side of the antennas center line). The upper and lower sections of the DB4E being the core components of two DB2E antennas. The wire diameter and spacing of the balanced lines is chosen to provide a characteristic impedance appropriate for use as a quarter wave matching line.
http://en.wikipedia.org/wiki/Quarter...ce_transformer The net result is that both the Antennas Direct designs I've seen (and presumably most of their other original designs) appear to present a nominal 300Ω balanced antenna output to the input of the matching network.
The matching networks on both antennas I have in hand are micro strip lines built from what appears to be glass/resin PCB material.
http://en.wikipedia.org/wiki/Microstrip The function of strip-line network is to transform the 300Ω balanced input to a 75Ω unbalanced output. There is very little ohmic loss in the small amount of copper of the strip-line, as a result this type of matching network is known for low insertion loss. (The matching network can be thought of as analogous to a car's transmission... allowing efficient coupling between an engine turning at high speed and the tires turning at low speed. To take the analogy further, a transmission designed with gears will be more efficient than one using pulleys and belts, given that gears slip less and have less friction than belts and pulleys. The micro strip being like the gear design and wire wound transformers being like belt and pulleys.)
So as I look at the photos of the DB8E, I see two DB4E's ganged, interconnected by two equal length coax cables and a 'black-box'. I'm left with two 'guesstimates' of what they have done...
1) The two cables are custom impedance (intended to act similarly to quarter-wave-transformers) and the 'black-box' is a simple set of connectors that parallel the two with one output.
2) The two cables are common 75Ω coax and the 'black-box' is a low loss micro-strip network. (Based on what I've seen from the engineering at AD, I would bet this is the more likely of the two possibilities.)
To the question of multipath, I don't expect any antenna design to overcome multipath by any means other than 'directionality'. In other words, the more directional the antenna is, the less off bore-site signals will be received... And if the interfering multipath(s) are arriving at the antenna from off bore-site direction(s) then the more directional design will help combat the multipath problem.
If you configure the DB8E to be a less directional array, you give up some of your multipath rejection. I don't see that as a reason to be critical of the design... That's physics.
A reversed splitter is a not an impedance matching devise, it's a power dividing network. As such, it will present more insertion loss than the micro-strip-line matching networks used by AD. When working with power transmission, impedance matching translates to efficient (low-loss) coupling of power. A matching network can have several ports, (the phasing lines running from each 'bow-tie' element on the typical panel antenna design for example). As power travels along the line, it will be adsorbed little by ports with mismatched impedance and absorbed efficiently by a port with a matching impedance. As a result, a correctly designed network will reduce the amount of power retransmitted by ganged elements.
http://www.antennasdirect.com/cmss_f...s/DB8E-TDS.pdf