There’s a danger in that arrangement of having the worst of both worlds. Perhaps I need to explain the problem.
The data to and from the sensors is digital, in the form of pulses. These travel in cables at a finite speed, lets say about 1 foot per nanosecond. When they come to the end of the cable, unless a very specific condition is met (the cable is terminated in its characteristic impedance), the pulse gets reflected. Let’s say t6 is sending its data. When the first pulse gets to the junction, some will go into the emonPi and be recorded, some will go the other way towards t2, get to the end and be reflected back, and reach the emonPi weaker, but it will still get there. There it interferes with later pulses from t6 that are just arriving at the emonPi. That’s what causes the problems. You might get away with what you’ve drawn, but then again you might not. If feasible, just one cable looped around all the sensors (all 12) would likely be better. There are a number of methods of lowering the possibility of reflections causing trouble, this https://pdfserv.maximintegrated.com/en/an/AN148.pdf has a lot of information but it’s quite technical. A point to bear in mind: Maxim assumes you are driving the bus in the optimal way at the controller end. The emonPi is not like that, all it has is a plain pull-up resistor.
I don’t see why not. You’ll probably need to have the adhesive-lined type of heat-shrink to get a good waterproof seal. The glue melts as the sleeve shrinks and forms a very good seal. You could easily twist and solder 3 wires together (cut ends all pointing the same way), slip the glued sleeve over everything and shrink it, then while the glue is still molten, crimp the open end flat while the glue hardens to seal it. Note the spur to the sensor should be as short as possible.
That’s already answered here: DS18B20 reliability considerations - #7 by pb66