My intention is a switch, so I guess an optocoupler would be okay. (most of optocouplers I've seen are rated for small voltage and current like 5V-12V though - I'll have to look for a faster one.)

If I used the "dimmer" solution though, wouldn't increasing the voltage amplitude (i.e. in a way that would distort the first 1-2 cycles of the 50HZ AC) just in a way like pulling the dimmer gradually up from 0 to 100% over some 200ms reduce the negative effect? (so instead of detecting zero crossing, I just allow no more than 1%/2ms change in software) reduce the EMI? If I'm not mistaken, it's caused by rapid spike resulting from switching from 0 to max current in much less than the 50HZ 1/4 cycle time, and this way - most devices should tollerate the slower rise than the rapid spike.
(sorry if I didn't make myself clear, if so, just tell me and I'll try to draw my idea.)