This phenomenon is caused by the final circuit cable capacitance. It is more common in conventionally wired two way and intermediate switching circuits due to the wide spread use of 3+E cable and the extra parallel capacitance that creates. LED lamps are particularly sensitive to this because they require very little current to drive the lamp.
The mains LED lamp is essentially a number of LEDs driven by a constant current driver powered by a mains rectifier and smoothing capacitor.
Even though the live feed to the LED lamp is switched OFF the cable capacitance allows leakage current to flow to the lamp slowly charging the smoothing capacitor in the lamp until the potential is sufficient for the constant current driver to "fire" the LEDs. This discharges the smoothing capacitor, the constant current driver stops, the LEDs extinguish and the process of charging starts again ... if conditions are suitable this continues indefinitely.
If the charge/discharge cycle (repetition rate) is short, the LEDs appear to be 'ON' continuously, however the energy is limited so the LEDs are only barely lit.
If more lamps are added to the final circuit the loading on the cable capacitance can be such that the individual smoothing capacitors of some or all of the lamps do not reach "firing" potential. Or you may find that only one or a few lamps on a circuit are effected due to the distribution of the cable capacitance and the manufacturing tolerances of the lamps. This can be quite disconcerting the first time you encounter it.
A tungsten filament lamp doesn't suffer this problem because the filament is permanently loading the cable capacitance and any acquired charge is unable to overcome the thermal inertia.
A simple solution is to permanently discharge the cable capacitance by wiring two 160K ohm, 1/4W, resistors rated at >=250V, in series across L and N on the final circuit, it is good electronics practice to do this as close to the effected lamp as possible, if this is done across the switched L and N at the lamp holder terminals it allows simple switching out of circuit (or disconnection) for testing. Do not try to use a single 330K ohm resistor because this may fail due to voltage breakdown.
If the final circuit has a large number of switched lamps it may be pragmatic to connect the series resistors between permanent L and N at a convenient point e.g. at the DB to reduce the number of resistors required, however the number of points that will require treatment will depend on how the cable capacitance is distributed on the final circuit. Additional points particularly where 3+E is used for 2 way and intermediate switching may well require additional pairs of resistors.
If this solution is implemented a notice should be placed at the DB stating which final circuits are effected and how they have been modified.
