I tried it with the TSOP4136 too, on a Itron CL200 meter in the USA. It never worked well and consistently. I thought it was a noise issue too.
After some trial and error, I discovered that the TSOP4136 is totally unsuitable this purpose. It detects modulated signals, and I am not even sure that my electric meter was outputting modulated signals. Their technical specs for the meter don't mention anything about it.
On a hunch, I tried just an simple infrared detector from Sparkfun, SEN-00241 https://www.sparkfun.com/products/241 That did the trick. Now I detect the pulses extremely reliably, to every single watt-hour to be able to predict when the meter will flip over to the next kWh digit. The only possible issue is that the sensor is certainly sensitive to ambient illumination, so I have to shield it. The ambient infrared tends to pull the signal low, so to get a clear HIGH and LOW signal, I route it through a comparator, comparing it against a reference voltage of about 3.5 V (arrived at by trial and error). I don't have a scope to verify the purity of my signal, but I was able to measure the pulse width to be consistenly 10.025+/- 8 milliseconds.