I assume you mean the 4K7 pull-up resistor on the OneWire bus? If you’re running your sensors in parasitic power mode then (at least with the DS18B20) you need to implement an active pull-up. You need the weak 4K7 pull-up at some stages of the transaction, and you need to drive the bus hard high during other stages in order to feed the sensors enough current during the conversion. The datasheet has more details.
In any case, I believe the OEM set-up is 3-wire 1-wire, i.e. +5V, 1-Wire and GND. 5V gets fed to all the sensors on a separate wire so there is no parasitic power in that case.
I think that could well be what’s happening, not because of the 4K7 pull-up (which isn’t involved in the supply in non-parasitic operations) but for the reason Robert pointed out above. The OP hasn’t answered Robert’s question yet, but if his EmonTX is being powered by the half-wave rectifier hanging off the AC-sensing transformer then he may well be putting way too big a load on the 5V rail with 9 simultaneous conversions. Here’s a picture of 6 simultaneous conversions using a 5V Arduino (I don’t have any OEM hardware):
The pink channel is just a trigger channel to indicate when the ConvertT command is being sent. That command is going to all 6 sensors on the bus because of a prior SkipRom command (not shown). The blue channel is the 1-Wire bus and you can see the 0x44 being sent LSB first.
The interesting one is the green channel. It’s the 5V supply feeding the 6 sensors, but after I’ve passed it through a 100R to give me a convenient way of measuring the current. The only thing downstream of the 100R is the 6 sensors. You can see pretty much immediately after all those sensors see the ConvertT command, the 5V supply sags from 4.96V down to 4.3075V. 652.5mV across a 100R represents a current draw of 6.525mA or about 1.1mA per sensor (consistent with the datasheet which says typical 1mA, max 1.5mA).
If that is the source of the problem then a possible fix might be to tweak the code to do one conversion at a time (call requestTemperaturesByAddress() or requestTemperaturesByIndex() instead of requestTemperatures()).
Actually, the more sensors (and cable) you add to the bus, the more capacitance you add. Here’s a picture of one of my 6 sensors trying to transmit a ‘1’ bit by doing nothing in response to the master’s bus pulse:
The master drives the bus low for 3.5 usecs, then tri-states it and then checks the bus state 9.5 usecs later to see if it’s high. So that right cursor, 13 usecs after the initial master falling edge represents the time at which the master samples the bus. In this case the 4K7 R had pulled it back high in time, but the more devices you add, the longer it takes, so if that’s the problem here then making the 4K7 stronger might help.
