OK, sorry I did not realise it was Dallas one-wire, although you are using it with separate 3V3 which is a very good idea.
As rpdom said, the star topology will be the worst, a daisy chained the best.
I had a look at the Dallas datasheet, the device is OK with sinking 4mA. This means they will be OK down to a 1k pull-up resistor at a 3V3 supply (3.3V / 1k = 3.3mA). Your problem will most likely be either be due to reflections or slow rise times on the data line. If it were my system, I would try a 2k2 resistor instead of the standard 4k7 and see if it's more reliable. This will improve the rise time of the data signal. As I said, 1k should also be safe, although it's much lower than the standard 4k7.
To really find out what's going on, you need an oscilloscope on the data line at RPi and also at the Dallas, but that's probably advanced debugging. If you Google around, people have best success the cat5 or twisted pair telephone cable. If you are using twisted pair, run the ground and data lines as a twisted pair.
"The time has come," the Walrus said,
"To talk of many things..."